This study aims to resolve a paradox. Experiments measuring alpha band power report an event related decrease (desynchronization) in alpha activity, whereas those measuring evoked alpha report synchronization. During a recognition memory task with human subjects, we measured the evoked (phase locked) and induced (not phase locked) alpha response. The findings reveal that evoked alpha is due to a transient phase locking (at about 100-200 ms poststimulus) of three alpha sub-bands which can be observed only at parieto-occipital sites. In contrast, induced alpha shows a widespread pattern of desynchronization at most recording sites. Thus, opposite alpha responses occur within similar time windows. Evoked alpha synchronization may reflect cortical inhibition which serves to increase the signal to noise ratio for activation processes following immediately later.