In oddball paradigms, infrequent stimuli elicit larger P300 event related potentials (ERPs) than frequent ones. One hypothesis is that P300 modulations reflect the degree of “surprise” associated with unexpected stimuli. That is the P300 represents how unlikely the stimulus is and this signal is then used to update the observer’s expectations. It could be hypothesized that P300 is modulated by any factor affecting an observer’s expectations, not only target probability. Alternatively, the P300 may reflect an evaluative process engaged whenever a discrepancy between task context and sensory inputs arises, irrespective of the latter probability. In previous ERP studies, stimulus probability was often the only determinant of task set confounding the effects of stimulus probability and set stimulus discrepancy. In this study, we used a speeded luminance detection task. The target was preceded by a central cue that predicted its location. The probability that the target was valid, i.e. would appear at the cued location was manipulated by varying the reliability of the cue. Reaction times were modulated by probabilistic expectations based on cue reliability and target validity while P300 was affected by target validity only. We conclude that increased P300 amplitude reflects primarily breaches of non-probabilistic expectations, rather than target probability.