Response times (RT) curves were obtained from subjects who were instructed to search for following or preceding letters in the alphabet with respect to given stimulus letters. To describe the shape of these curves, Klahr, Chase, and Lovelace (1983) proposed a search model. They assumed that the alphabet was internally represented by a number of chunks of a few letters. Furthermore, they assumed that serial searchs occurred along two hierarchical levels: (a) a search for the correct chunk (RT increases across the alphabet) and (b) a search of the letters within the chunk (RT increases within chunks). Their model predicted an ascending sawtooth-shaped RT curve. However, according to our analysis of their data, the RT curves did not show the expected sawtooth shapes. Besides, the assumptions underlying the model were, in our view, implausible. In order to find out whether the lack of empirical support for the model happened to be an artifact of averaging the data, we replicated their experiment. We found that individual RT curves did not show the shape characteristics that were predicted by the model of Klahr et al. either. The data rather supported the hypothesis that the access of letters was direct without any hierarchy, and that the retrieval of successive letters was determined by the variable associations between subsequent letters.