If the contextual similarity between learning and recall within a single trial in a short-term memory (STM) paradigm is varied, recall varies proportionately. This context effect was demonstrated using variations of the Peterson-Peterson paradigm for both aurally and visually presented material, verbal and arithmetic context, and within-and between-5s designs. Experiments were conducted to discover whether the context effect was due to differential ability to rehearse in the recall interval, differing intertrial intervals, and different amounts of activity on 6"s part during a trial. None of these hypotheses were supported by the data. An analysis of intrusion errors fails to support the hypothesis that the context effect is due to proactive interference. An explanation of the effect on the basis of stimulus generalization is proposed.

Manipulation of the context in a short-term memory paradigm produces changes in the ability to recognize the same material from long-term memory 24 hr. later. If immediate recall is accurate, later recognition is improved if this recall is conducted with the same context as occurred at learning. If immediate recall is completely inaccurate, later recognition is improved if this recall is conducted with different context than was present at learning. Short-term recall did not need to be accurate to transfer the learned nonsense trigrams to long-term memory. Manipulation of context 24 hr. after learning had no effect on recognition. Results are discussed in terms of the Waugh and Norman memory model, Tulving's encoding specificity hypothesis, and interference theory.