Dose effects of secobarbital on free recall and recognition of words were examined in levels-of-processing paradigms. Secobarbital significantly impaired recall, even when initial item processing was guided with appropriate orienting tasks. However, when processing was guided both at input (with orienting tasks) and at retrieval (with recognition testing), secobarbital-related retention deficits reported in conventional 'learn these words' experiments were not found. Levels-of-processing theory suggests at least two possible mediators for retention deficits associated with drug intoxication, a production deficit at encoding or reduced processing capacity. Neither of these hypotheses can account for the data. A third hypothesis suggests that the intoxicated subject adopts a cautious response-decision strategy, thus failing to emit available items in recall testing or to endorse correct items on recognition. The results of signal detection analysis offered no support for this hypothesis. Considered together with earlier studies, the data indicate that when processing is guided only at retrieval, secobarbital produces both recall and recognition decrements. On the other hand, when processing is guided only at input, the drug produces recall deficits. Apparently, when secobarbital-intoxicated subjects are left to generate their own responses at retrieval, they fail to undertake optimal retrieval strategies. When their processing is guided both at input and retrieval, their drug-induced retention decrements can be corrected.