Four experiments demonstrated that visual search can be decomposed into two components: one consisting of skills shared with memory search and the other consisting of skills not shared with memory search. A training–transfer paradigm was used to test for transfer from memory search to visual search and vice versa. When the same targets and distractors were used in training and transfer, visual search practice completely trained memory search, but memory search practice only partially trained visual search. Learning on both the shared and the private components of visual search benefited more from item-specific training than from nonspecific training. The relationship between the components and some theorized models of visual search are discussed, particularly in terms of prioritization learning.