The effects of different masking conditions on identification of face gender and expression at different target durations (17-119 ms time range) were studied in two experiments. In Experiments 1a and 1b, the effects of face masks were compared against those of noise masks (scrambled face stimuli) and a control, no-mask condition. Significant masking by noise masks was only found at the shortest target duration (17 ms). Effective masking by face masks was observed over a slightly longer time window in the gender than in the expression task (17-85 and 17-51 ms respectively). Moreover, clearly differentiated effects of face and noise masks were observed only in the expression task. In Experiments 2a and 2b, faces quantized with different sampling sizes were used as masks. A graded effect of sampling size was observed in the expression task, with those masks that preserved more facial information exerting stronger masking. However, all masks were equally effective in the gender task. These results demonstrate an interaction between masking and task demands, suggesting that different processing mechanisms may underlie identification of different properties of faces. An interpretation is offered in terms of the relative role of configural and feature processing in expression and gender identification.