Two basic types of experimental designs are available to measure the brain indices of successful recognition memory. Both designs have in common that brain activity elicited by correctly recognized studied (old) items is compared to brain activity elicited by correctly rejected new items. In blocked designs, the two item types are presented in separate lists. In random-mixed designs, on the other hand, they are randomly intermixed within the same list. Early studies reported that in random-mixed designs, brain activity can be affected by the sequence in which different types of items are presented. Here, we investigated to what extent such sequence effects also occur in random-mixed designs of recognition memory. Event-related potentials (ERPs) were obtained while subjects made recognition judgments on pseudorandom sequences of visually presented new and old words. Changes in the sequence from one item-type to another (change from old to new and from new to old) were associated with a prominent P300 potential for new words following old words but not vice versa. Subjects with larger P300 potentials made less false alarms to such new words. Successive presentations of the same item-type (sequence 'blocks': three old words in a row and three new words in row) were associated with a left parietal positive slow shift for successive old words but not for successive new words. The data suggest that new words require extensive context-updating if they are presented within the context generated by preceding old words. Due to these sequence effects, the present data question the assumption that random-mixed designs are generally favorable over block designs.