The pattern of sound energy spread across frequency and time (spectro-temporal modulation) is a crucial stimulus cue for sound identification. Improvements in the sensitivity to this modulation could potentially aid performance on numerous real-world tasks, yet it is unknown how practice influences this sensitivity. To investigate this issue, normal-hearing adults (n=8) were trained ∼1 h/day for 7 days to discriminate between noises with the same spectro-temporal modulation drifting upward in audio frequency but with different modulation depths. Performance on the trained condition and on four untrained conditions was examined both before and after the training phase. Depth-discrimination thresholds improved significantly on the trained condition and on an untrained condition with a downward-drifting modulation. Discrimination thresholds also improved in subsets of listeners on untrained isolated component modulations (spectral or temporal) of the trained spectro-temporal modulation. Finally, the ability to detect the trained spectro-temporal modulation worsened significantly in proportion to the amount of improvement on the trained discrimination task. These data suggest that training on depth discrimination may be a means of improving sensitivity to spectro-temporal modulation depth in real-world stimuli, but potentially at the cost of the ability to detect these modulations. [Work supported by NIH/NIDCD]