This study extends the unidimensional view of review helpfulness into a bidimensional one by conceptualizing review helpfulness and unhelpfulness as two distinct constructs, and draws on the lens of social learning theory to examine the effects of review helpfulness and unhelpfulness on subsequent voting behavior. We innovatively adopt the latent change score modeling to analyze a longitudinal review dataset from Amazon. com. Our empirical results indicate that prior helpfulness of a review will be positively related to its subsequent increase in helpfulness and unhelpfulness, while prior unhelpfulness of a review will be negatively related to its subsequent increase in helpfulness and unhelpfulness. The results also indicate that, at the aggregate level, the increases in review helpfulness and unhelpfulness decelerate over time. This research contributes to the literature by unravelling the dynamic reciprocal relationship between review helpfulness and unhelpfulness and sheds light on the design of review voting systems.