The quality of service (QoS) of cloud services change frequently over time. Existing service recommendation approaches either ignore this property or address it inadequately, leading to ineffective service recommendation. In this paper, we propose a time-aware service recommendation (taSR) approach to address this issue. We first develop a novel similarity-enhanced collaborative filtering (CF) approach to capture the time feature of user similarity and address the data sparsity in the existing PITs (point in time). We then apply autoregressive integrated moving average model (ARIMA) to predict the QoS values in the future PIT under QoS instantaneity. We evaluate the proposed approach and compare it to the state-of-the-art. Our experimental results show that taSR achieves significant performance improvements over existing approaches.