A 500-Hz tonal signal was masked by a continuous tone, both from the same oscillator. A two-interval-forced-choice procedure was used to determine the level required for 80% correct. The masker level was 50 dB SPL, and the signal was always in phase with the masker. Signal durations from 20 msec to 2 sec were employed. The results showed a slope of approximately 3 dB per doubling over the entire range from 20 msec to 1 sec. Detection at 2 sec tended to be worse than at 1 sec. The slope of 3 dB per doubling is much steeper than has commonly been found with noise maskers, especially at the longer durations. Data from an electrical model are discussed.