The purpose of the present study was to investigate on what experimental condition visual imagery ability could successfully predict the subject's performance in memory task. To attain this goal the imagery ability of individual subject was measured by the use of VVIQ, the following three factors were varied in associative memory task. The first factor was the difficulty in associating stimulus term with response one. It was controled by varying the number of response terms, one (in Exp. 1) or two (in Exp. 2). The second was the kind of memory task such as short or long term recall. The former was an immediate recall of response items, while the other was an unexpected free recall in which the subjects were asked to recall both stimulus and response items 30 minutes later after short term recall. The third was the image-arousing p otential of material. That potential was defined as the integrated imagery value (high or low) which was the vividness of imagery constructed of pair (in Exp. 1) or triplet (in Exp. 2) of words. Prior to two experiments 30 undergraduate sudents completed the VVIQ, and were divided into two homogeneous groups of 15 subjects on the basis of their VVIQ scores. The one group took part in Exp. 1, and the other in Exp. 2. The subjects in Exp. 1 and 2 were together instructed to relate the stimulus term with the response one and construct the integrated imagery. The subjects had no prior experience of imagery experiment, and their VVIQ scores were unknown to the experimenter to exclude the influence of demand characteristics. After each experiment the 7 lowest and the 7 highest scores of VVIQ were selected as “good” and “poor” imagers respectively. The basic data were the number of total words recalled by each subject. In each experiment a 2×2×2 (imagery ability, integrated imagery value, and trial) and a 2×2 (imagery ability and integrated imagery value) analysis of variance were applied to short and long term recall data respectively. The integrated imagery value and the trial were repeated measures. Main results and conclusion were as fol lows: a) The high imagery material (H set) was superior in the number of recalled words to the low imagery material (L set) on both short and long term recall in Exp. 1 and 2. b) The short term recall was improved on thesecond trial over the first trial in both Exp. 1 and 2 (FIG. 1 and 3). c) The imagery ability x integrated imagery value interaction was significant on long term recall in Exp. 2, indicating that H set was remembered better than L set in poor imagers and the difference between H and L sets was not found in good imagers (FIG. 4). This result suggested that good imagers could be successful in equally transferring H and L sets to long term store (LTS) by sufficient imaginal coding, and poor imagers failed in transferring L set to LTS in contrast to H set. On short term recall in Exp. 2, the same interaction was found at the second trial (FIG. 3). In the secondtrial most of recalled words were probably retrieved from LTS, because the materials to be learned were identical with those of the first trial. Consequently the interaction may be interpreted as caused by the same mechanism in long term recall. From the above results, it is implied that the imagery ability might be successful in predicting the subject's performance in recall of L set constructed of triplets of words from LTS. That implication supported the following result. d) There was a positive significant correlation between the subject's imagery ability and the number of words recalled from L set on long term recall in Exp. 2 (FIG. 5).