In addition to testing the echo duration, the Simrad EK500 split-beam echo-sounder uses the between-sample variation in signal phase to isolate echoes from single fish. The performance of these discriminators was tank-tested on a 38-kHz EK500 sounder, using spheres as targets. It was found that for typical settings of the allowable phase variation and echo duration: (1) there was a pronounced bias against accepting weaker targets; and (2) multiple echoes from targets as far as 700 mm apart were falsely accepted as single echoes, particularly if the individual echoes happened to interfere constructively. These effects will all result in target strengths being over-estimated, by an amount which could be large in many cases and which cannot be corrected for. We conclude that, except in cases where the fish are widely separated and differ comparatively little in target strength, the single fish discriminators used in the EK500 (software Version 4.01) are unreliable, and should be used with great circumspection. Possible improvements to the phase algorithm are suggested, although it is considered unlikely that these will eliminate all biases in the method.