Two phantoms for assessing the performance of ultrasound scanners regarding detectability of small focal lesions are described. The spherical simulated lesions in the phantoms have diameters of 2.4, 3.0, and 4.0 mm and backscatter coefficients which are 16, 9, and 6 dB below that of the surrounding tissue-mimicking material, the latter simulating normal tissue such as liver. Random positioning and relatively large numbers of lesions deal with the statistical problem related to mistaking a normal textural fluctuation, characteristic of ultrasound images of parenchymal tissue, to be a small lesion. All lesions have ultrasonic properties that are the same as those of the surrounding material except for the backscatter coefficient; therefore, artifacts resulting from attenuation, reverberation, and/or refraction are absent, and detectability of deeper lesions is not influenced by the presence of more proximal ones. A method is outlined for using the phantoms to determine bounding proximal and distal depths of resolution, or detectability, of focal lesions of a given object contrast and size. Tests for observer dependence involving five subjects resulted in standard deviations in the distal and proximal depths of less than 0.5 cm. A repeat of the observer dependence tests 18 months later showed little change in the results. In addition, results obtained by two experienced observers are also given for a variety of specific scanners, scanning heads, and scanning parameters.