Previous research has identified two subgroups of patients with Alzheimer's disease (AD) based on performance discrepancies on semantic and visual-constructional measures: Left AD (LAD) and Right AD (RAD). In this study, verbal fluency performances (Animal Fluency [AF] and Letter Fluency [FAS]) of these two subgroups were examined. It was hypothesized that LAD patients would perform worse on AF compared to FAS, due to an underlying breakdown of left-hemisphere semantic networks. On the other hand, the RAD group, which theoretically has a relatively preserved semantic system, yet difficulties retrieving overlearned information, was not expected to differ on the two fluency tasks. These predictions were based on the notion that the AF task requires intact retrieval and semantic processes, whereas the FAS task is reliant on retrieval processes alone. Patients were classified into subgroups on the basis of performance discrepancies on the Boston Naming Test (BNT) and Copy tasks: LAD (BNT < Copy); RAD (BNT > Copy). A split-plot ANOVA using demographically corrected standard T-scores revealed a significant main effect for fluency task, and a significant subgroup x fluency task interaction. LAD patients performed poorer on AF compared to FAS; there was no fluency task difference for the RAD group. Analysis of within-subcategory response clustering on AF revealed more instances of serial subclass exemplar responses by RAD members. These results support the loss theory in explaining the semantic deficit of LAD, and suggest that retrieval difficulties underlie the fluency problems of RAD.