ABSTRACT A United States Environmental Protection Agency (EPA) research program evaluated the current performance of commercially available volumetric test methods for the detection of small leaks in underground gasoline storage tanks. The evaluations were performed at the EPA Risk Reduction Engineering Laboratory's Underground Storage Tank Test Apparatus in Edison, New Jersey. The methodology used for evaluation made it possible to determine and resolve most of the technological and engineering issues associated with volumetric leak detection, as well as to define the current practice of commercially available test methods. The approach used (1) experimentally validated models of the important sources of ambient noise that affect volume changes in nonleaking and leaking tanks, (2) a large data base of product-temperature changes that result from the delivery of product to a tank at a different temperature than the product in the tank, and (3) a mathematical model of each test method to estimate the performance of that method. The test-method model includes the instrumentation noise, the configuration of the sensors, the test protocol, the data analysis algorithms, and the detection criterion. Twenty-five commercially available volumetric leak detection systems were evaluated. The leak rate measurable by these systems ranged from 0.26 to 6.78 L/h (0.07 to 1.79 gal/h), with a probability of detection of 0.95 and a probability of false alarm of 0.05. Five methods achieved a performance between 0.19 L/h (0.05 gal/h) and 0.57 L/h (0.15 gal/h). Only one method was able to detect leaks less than 0.57 L/h (0.15 gal/h) if the probability of detection was increased to 0.99 and the probability of false alarm was decreased to 0.01. The measurable leak rates ranged from 0.45 to 12.94 L/h (0.12 to 3.42 gal/h) with these more stringent detection and false alarm parameters. The performance of the methods evaluated was primarily limited by test protocol, operational sensor configuration, data analysis, and calibration, rather than by hardware. The experimental analysis and model calculations suggested that substantial performance improvements can be realized by making procedural changes. With modifications, it is estimated that more than 60 percent of the methods should be able to achieve a probability of detection of 0.99 and a probability of false alarm of 0.01 for leak rates between 0.19 L/h (0.05 gal/h) and 0.56 L/h (0.15 gal/h), and 100 percent should be able to achieve this performance for leak rates of approximately 0.76 L/h (0.20 gal/h).