Psychometric functions were measured for spectral-shape discrimination (profile analysis) with the signal frequency either fixed, what we call the signal-known condition, or randomly varied, what we call the signal-unknown condition. The functions were obtained using an adaptive, up-down procedure. In the signal-unknown condition, independent tracks for each signal frequency were interleaved within the same block of trials. The mean slope of the psychometric function [k in log d' = k log delta L+C, where delta L = 20 log(1 + delta p/p)] was 1.13 for the signal-known conditions and 1.19 for the signal-unknown condition. The mean signal-to-standard ratio at threshold (Pc = 79.4%) obtained in the signal-unknown condition was 3 to 4 dB higher than that obtained in the signal-known conditions. The psychometric functions of the ideal observer were derived for both the signal-known and signal-unknown conditions. A comparison of the measured and derived psychometric functions suggests that the internal noise component that is statistically independent across frequency channels contributes minimally to the total (internal and external) noise sources that are involved in the decision process.