<p>Accurate urban flow prediction plays a crucial role in transportation management, as it enables optimized resource allocation and improved traffic efficiency. Although current methods have made advances, they still encounter challenges such as high computational overhead and the risk of overfitting complex models. To tackle these issues, we introduce the partial channel connection to urban flow prediction, aiming to reduce complexity while keeping competitive performance. The partial channel connection selectively engages a specific subset of channels for operations in a single step, while preserving the identity mapping for the remaining channels. This approach ensures comprehensive training of all channels throughout the entirety of the training process and mitigates computational overheads at each step. In this paper, we apply the partial channel connection across various spatial and temporal encoders, undertaking a thorough investigation into their predictive accuracy. Based on these insights, we design a model named PartialST for urban flow prediction, which effectively captures the temporal and spatial correlations. We evaluate PartialST through comparative experiments against other state-of-the-art methods on two real-world datasets. The results demonstrate not only the superior performance of our model over other comparative models but also the effectiveness of the partial channel connection approach.</p>