Attention to sensory stimuli is never uniformly distributed. We tested whether time-based and feature-based aspects of sensory attention interact in facilitating the detection of new stimuli in a stream. We recorded behavioural and encephalographic (EEG) data while participants attended to repeating pure tones (standard tones) which unpredictably changed in feature (deviant tones). Participants responded more rapidly to deviant tones longer waited for (time or Hazard rate effect), as well as to deviant events carrying larger rather than smaller deviancy magnitudes (feature effect). However, there was no interaction of time and feature dimensions. Attention increased phase coherence more consistently for small deviancy signals rather than for large deviancy signals. The hazard rate was reflected by the suppression of Beta (14-24 Hz) power relative to Alpha (8-12 Hz) power. Time-based (when) and feature-based (what) aspects of sensory attention seem to partition distinct neural processes modulating separate behavioural components.