The data of Anderson and Burr [1985. Vision Research, 25, 1147-1154] on the
temporal-frequency (TF) specificity of noise maskers indicate that the eff
ect of TF masking is broad and varies across spatial frequency (SF) channel
s. One subtle but significant feature of the data is that the TF at which t
he effect of masking is maximal falls continuously as the test TF falls. Th
is continuous shift is hard to reconcile with models of detection in the li
terature that relate detection to the most sensitive filter, without resort
ing to a large number of temporal filters. We developed a new model, which
relies on only three temporal filters and posits that detection is the resu
lt of a threshold decision based on the compound Bayesian probability of al
l filter responses, not just the most sensitive filter. (C) 2000 Elsevier S
cience Ltd. All rights reserved.