This paper focuses on the review voting in online communities, which allows users to express their own opinions in terms of User-generated Content (UGC). However, the sustainable development of online communities is likely to be affected by the social influence of UGC. In this paper, we study the so-called crowd intelligence paradox of review voting in online communities. The crowd intelligence paradox means that the quality of reviews is not highly connected with the increasing of review votes. This implies that a review with many votes is likely to be of low quality, and a review with few votes is likely to be of high quality. The crowd intelligence paradox existing in online communities inhibits users’ wishes of participating in social networks and may impact the sustainable development of online communities. Aiming to demonstrate the existence of the crowd intelligence paradox in online communities, we first analyzed a large set of reviews crawled from Net Ease Cloud Music, which is one of the most popular online communities in China. The maximum likelihood (ML) and the hierarchical regression approaches are used in this step. Then, we construct a new research model called the Voting Adoption Model (VAM) to study how different factors impact the crowd intelligence paradox in online communities. Particularly, we propose six hypotheses based on the VAM model and conduct experiments based on the measurement model and the structural model to evaluate the hypotheses. The results show that the quality of reviews is not influential to review votes, and the hot-site attribute is a dominant factor influencing review voting. In addition, the variables of the VAM model, including information credibility, perceived ease of use, and social influence have significant impacts on review voting. Finally, based on the empirical study, we present some research implications and suggestions for online communities to realize healthy and sustainable development in the future.