
The gallons-per-ton rule is a compressed form of the buffer run-time formula V = (24 / delta-T) x t, where 24 is the GPM per ton at a given temperature lift and t is the target run-time in minutes; the published range is wide because sources assume different run-times and system temperature differences, and the rule should be replaced by that calculation whenever low delta-T or deep chiller turndown is in play.
If you have specified a chilled water plant you have seen the rule: size the buffer tank at some number of gallons per ton of refrigeration. It appears in guides, in sales literature and in the back of estimating spreadsheets, and the value quoted ranges from about one to several gallons per ton depending on which source you open. The rule is useful as a sanity check and dangerous as a specification, because it hides the two inputs that decide whether it applies to your plant.
This article derives the rule from first principles so you can see what it assumes, explains why the published range is wide, names the two conditions that break it, and gives the run-time calculation that should replace it when either condition is present. The fluid relations are drawn from the ASHRAE Handbook; the chiller performance basis is AHRI 550/590 for water-cooling chillers and the reference data in CIBSE Guide B. The reader is the mechanical engineer asked to defend or dispute a buffer volume that someone pulled from a rule of thumb.
The short version: the gallons-per-ton number is a compressed formula, and once you expand it you can decide whether the assumptions behind it match your plant.
The rule states a buffer tank volume proportional to chiller capacity expressed in tons: so many gallons of tank per ton of refrigeration. It is a single number meant to cover every chilled water plant, and it survives because it is easy to apply during early estimating when the real inputs are not yet known.
Its origin is the run-time argument: a buffer tank exists so the chiller does not short-cycle, and the volume needed scales with how much cooling the plant must hold steady while the load swings. Expressed per ton, that volume becomes a ratio, and over decades of practice the ratio settled into a published range. The range is the tell: a true physical constant would not vary by a factor of several between reputable sources, so the number must be hiding assumptions rather than stating a law.
The published range is wide because the rule folds two plant-specific inputs into one constant, and different authors fold them differently. The first is the target run-time the buffer is meant to guarantee, measured in minutes of chiller operation or of pump flow. The second is the system temperature difference the plant actually runs across, which sets how many gallons of water represent a given amount of cooling.
A source that assumes a long run-time and a narrow system delta-T produces a large gallons-per-ton value; one that assumes a short run-time and a wide delta-T produces a small one. Neither is wrong in the abstract; each is simply the formula evaluated at its own defaults. The variation is not noise, it is the hidden inputs surfacing, which is why the rule must be expanded before it is trusted on a specific plant.
The two inputs the rule hides are, first, the target run-time t and, second, the system temperature difference delta-T_sys. Both enter the sizing through basic fluid arithmetic. In I-P units a ton of refrigeration is 12,000 Btu/h, and the water flow that carries it is GPM = 24 x tons / delta-T_sys, so the gallons per ton of flow is simply 24 / delta-T_sys.
The buffer volume per ton is then that flow multiplied by the target run-time in minutes: V_per_ton[gal] = (24 / delta-T_sys) x t[min]. With a 10 degF (5.6 K) system lift and a one-minute target the rule gives about 2.4 gallons per ton; halve the target run-time or widen the lift and the number moves with it. Every published gallons-per-ton value is this expression evaluated at some pair of t and delta-T that the author did not print.
Condition one that breaks the rule is the low delta-T syndrome: a plant where the return water arrives warmer than design, so the actual system temperature difference is smaller than the value the rule assumed. Because gallons per ton scales as 1 / delta-T_sys, a lower actual delta-T raises the volume the plant truly needs per ton, and the rule-of-thumb tank comes up short.
The failure shows up as the chiller cycling or the buffer temperature walking because the water mass is too small for the flow it is actually carrying. The fix is not a bigger rule-of-thumb multiplier but the calculation with the measured delta-T: measure the real lift, insert it into V = (24 / delta-T_actual) x t, and size to that. A plant known to run low delta-T should be sized on the low figure, not the catalogue average.
Condition two is deep chiller turndown, and it changes what the buffer is for. A chiller that modulates to a low fraction of capacity cycles less, so purely to prevent short cycling it needs less buffer than an on/off machine. Where a larger buffer is nonetheless specified alongside a deep-turndown chiller, the reason is flow decoupling rather than cycling: the tank reconciles the chiller's minimum primary flow with the variable secondary flow over the load swing.
That decoupling volume is set by flow and delta-T, not by the turndown label, so sizing the buffer from the turndown number alone is a mistake. Verify the buffer against the run-time calculation, and if the plant is primary-only add the primary-versus-secondary flow mismatch to the volume rather than scaling by turndown. The turndown tells you how the chiller behaves, not how big the tank must be.
The rule holds when the plant matches its hidden assumptions: a stable system delta-T near the design value and a chiller whose cycling the buffer is meant to tame. In that narrow case the published gallons-per-ton value, used with awareness of its basis, is a reasonable screen and a useful check on a detailed calculation.
It does not hold when the actual delta-T diverges from design, when the plant is primary-only and flow decoupling dominates, or when the chiller turndown changes the failure mode the tank must address. In any of those cases the rule-of-thumb volume is either too small or too large by an amount that scales with the mismatch, and the expanded calculation is the only defensible basis. Treat the rule as a check, never as the specification.
The calculation that replaces the rule starts from the duty: the buffer must hold the chiller's output for the target run-time at the actual system temperature difference. In I-P units the volume is V[gal] = (tons x 12,000 x t[h]) / (8.33 x delta-T_sys), or equivalently V_per_ton = (24 / delta-T_sys) x t[min], with delta-T_sys the measured lift and t the run-time the control sequence requires.
For a primary-only plant add the decoupling term: the tank volume must also cover the integrated difference between primary chiller flow and secondary load flow over the control cycle, evaluated at the same delta-T. The result is a volume with units of gallons that traces back to two numbers you can measure, not to a ratio you were quoted. Convert to the metric side with rho x cp if the project works in SI, but the structure of the formula is unchanged.
To check a vendor's proposed volume, expand it back into the rule and see what assumptions it implies. Divide the proposed gallons by the chiller tons to get the vendor's gallons-per-ton, then solve (24 / delta-T) x t = that value for the implied run-time at your system delta-T. If the implied run-time is far from what your control sequence needs, the volume is wrong for your plant.
Then run the full calculation at your measured delta-T and your required run-time, including any primary-only decoupling term, and compare. A vendor volume that matches the calculation at your actual inputs is sound; one that only matches a generic gallons-per-ton figure should be challenged. The check is cheap and it converts a mystery number into a documented decision.
Input | Relation | Effect on gal/ton |
Ton of refrigeration | 12,000 Btu/h | Defines capacity basis |
GPM per ton | 24 / delta-T_sys | Rises as lift falls |
System delta-T | Design lift, degF or K | Inverse on volume |
Run-time t | Target minutes | Linear on volume |
Gal/ton | (24 / delta-T) x t | Compressed rule value |
Measured delta-T | Actual return rise | Replaces design lift |
Condition | What to check | Action |
Stable delta-T | Actual lift near design | Rule may apply as screen |
Low delta-T | Return warmer than design | Re-size at measured lift |
Primary-only | Flow decoupling present | Add primary/secondary term |
Deep turndown | Chiller modulates low | Size from run-time, not turndown |
Vendor volume | Expand to implied run-time | Challenge if mismatch |
Calculation | At measured inputs | Use as specification |
Center Enamel (Shijiazhuang Zhengzhong Technology Co., Ltd.) designs and fabricates bolted steel storage tanks, welded steel tanks, fusion-bonded epoxy tanks, galvanized tanks, stainless steel vessels, pressure vessels, aluminum geodesic dome roofs and floating roofs. The company has more than 30 years of production experience and has supplied equipment to projects in more than 100 countries. Aluminum geodesic dome roofs are designed and fabricated to API 650 Appendix G, AWWA D108, the Aluminum Design Manual, ASCE 7 and IBC, and the company developed and exported a 72 m diameter aluminum dome roof in 2023. Engineering, fabrication, coating, quality control and site service are handled as one scope, which is why the drawings, the calculation basis and the delivered hardware can be reconciled against each other.
How many gallons per ton does a chilled water tank need?
The commonly published range is about one to several gallons per ton of refrigeration, but that spread is not uncertainty, it is the two hidden assumptions in the rule showing through. The number is a compressed form of V_per_ton = (24 / delta-T_sys) x t, evaluated at the run-time t and system temperature difference delta-T the author assumed. Read it as a screen, then confirm with the run-time calculation at your own delta-T and target run-time.
Is 10 gallons per ton the correct buffer tank rule?
Ten gallons per ton is not a universal constant; it is one point on the curve V_per_ton = (24 / delta-T_sys) x t. At a 10 degF (5.6 K) system lift and a one-minute run-time the formula gives about 2.4 gallons per ton, so ten implies either a much longer assumed run-time, a narrower delta-T, or a plant where flow decoupling dominates. Check which before adopting it; the correct value is the one your delta-T and run-time produce.
Why does the published range differ between sources?
The range differs because each source evaluates the same underlying formula, V_per_ton = (24 / delta-T_sys) x t, at its own assumed run-time and system temperature difference. A longer target run-time or a narrower delta-T yields a larger gallons-per-ton value, and vice versa. The variation is the hidden inputs surfacing, not disagreement about physics, so the right move is to expand the rule and apply your plant's actual inputs.
How does low delta-T affect the required volume?
Low delta-T raises the required volume because gallons per ton scales as 1 / delta-T_sys. When the return water arrives warmer than design, the actual system lift is smaller than the rule assumed, so each ton of cooling corresponds to more gallons of water at the reduced lift, and the rule-of-thumb tank comes up short and the chiller cycles. Re-size with the measured delta-T in V = (24 / delta-T_actual) x t rather than a catalogue average.
Why does a chiller with better turndown need more buffer volume?
Deeper turndown changes what the buffer is for rather than simply enlarging it. A chiller that modulates to a low fraction cycles less, so for pure anti-cycling protection it needs less buffer; where a larger buffer is still specified it is for flow decoupling in a primary-only plant, sized from primary-versus-secondary flow mismatch and delta-T, not from the turndown label. Verify the volume with the run-time calculation, not by scaling with turndown.
When should I ignore the gallons-per-ton rule?
Ignore the gallons-per-ton rule when the actual system delta-T diverges from design, when the plant is primary-only and flow decoupling dominates the volume, or when chiller turndown changes the failure mode the tank must address. In those cases the rule-of-thumb volume is off by an amount that scales with the mismatch. Use the run-time calculation at your measured delta-T and required run-time, adding the decoupling term for primary-only plants, and treat the rule only as a final check.
How do I check a vendor's proposed volume?
Expand the vendor volume back into the rule: divide proposed gallons by chiller tons to get their gallons-per-ton, then solve (24 / delta-T) x t = that value for the implied run-time at your system delta-T. If the implied run-time is far from your control requirement, challenge it. Then run the full calculation at your measured delta-T and required run-time, including any primary-only decoupling term, and accept the volume only if it matches your actual inputs.
The gallons-per-ton rule is not mystic, it is compressed: V_per_ton = (24 / delta-T_sys) x t, where the system temperature difference and the target run-time are the two inputs the published value hides. Once you expand it, the wide range between sources stops being a mystery and becomes the different assumptions those sources made.
Use the rule as a screen, but specify the buffer from the run-time calculation at your measured delta-T and your control sequence's required run-time, adding the primary-only decoupling term where it applies. Send us the chiller tons, the actual system lift and the run-time the plant needs, and we will return a buffer volume that traces to those numbers rather than to a ratio quoted from a spreadsheet.
Send the duty, the volume basis, the governing code and the site constraints, and our engineering team will come back with a specification that a supplier can price and a reviewer can approve. We will also tell you plainly when a different tank type or a different scope is the better answer.