Recently we reported that the direction of "invisible" coherent motion was sensitized through repeated presentation. This indicates that in some cases perceptual learning is formed without attention directed specifically at a presented feature (Watanabe, Náñez & Sasaki, 2001; Nature). Here, we examined which stage(s) of visual processing is/are involved in passive learning. The first experiment (psychophysics) consisted of a training stage and pre- and post-test stages. In each trial of training, we presented stochastic random-dot cinematograms (SRDCs) in which a global motion flow is perceived in numerous dots moving spatio-temporally randomly within a certain range of directions, as an irrelevant stimulus. The subjects performed a RSVP task and orientation indication tasks while viewing the SRDCs. The global motion flow direction was constant for each subject. In test stages we measured the discriminability of 9 directions and found performance improvement after training only in the directions within which local dots moved in the training stage, irrespective of the global motion direction. The results indicate that passive learning occurred only at the local motion stage. In the second experiment, fMRI procedures (Siemens 3T) were used to specify the cortical area for the local and global motion. Motion displays where dots moved in completely random directions for 16sec were followed by an SRDC for 16sec. In each block this procedure occurred four times within 90 deg ranges in SRDCs covering a total of 360deg. There were 80 blocks in total in the experiment. No significant difference in the amount of signals between the random display and SRDC conditions was found in V1, while signals were significantly higher in MT+ and V3a for the SDRC condition. The results suggest that local motion is mainly processed in V1 and global motion in higher stages. From the results of the two experiments, we concluded that V1 may be the neural locus of passive learning.