Subjects searched a simple three level hierarchical filing system to locate hypothetical documents. Search was guided only by a numerical index. The index was either (a) the probability that the document would be found by selecting one of the alternatives, (b) a frequency count of possible hits available by selecting one of the alternatives, or (c) a bar-graph of the likelihood that each of the alternatives would lead to the target document. The three indices were calculated using either (a) joint probabilities or (b) probabilities normalized across the alternatives at each level. An analysis of search patterns indicated that subjects tended to follow highest probability paths for the first traversal. If the target was not found, they engaged in a local search at the bottom of the hierarchy. If the target was still not found, they attempted to follow the second highest probability path down. Results indicated that subjects made the fewest deviations from the best path when probabilities were normalized. Furthermore, bar-graph displays uniformly resulted in the fewest deviations from the best path. When subjects redirected their search, they could either move up the hierarchy one level or restart at the top. Subjects with high spatial visualization ability were more likely to move up one level, while subjects with low spatial visualization were more likely to restart the search. Finally, if the target was not found after several attempts, search patterns began to reveal considerable redundant ra ndom search behavior and checking of paths. Techniques are suggested to reduce inefficient search patterns.