Accurate flow field estimation is crucial for the improvement of outdoor environmental quality, but computational fluid dynamics (CFD) based on the widely used Reynolds-averaged Navier–Stokes method has limitations in this regard. This study developed a turbulence modeling framework based on a convolutional neural network (CNN) to model turbulence in urban wind fields. The CNN model was trained by learning the Reynolds stress patterns and spatial correlations with the use of high-fidelity datasets. Next, the model was integrated into the CFD solver to generate accurate and continuous flow fields. The generalization capability of the proposed framework was initially demonstrated on the simplified benchmark configurations. The validated framework was then applied to case studies of urban wind environments to further assess its performance, and it was shown to be capable of delivering accurate predictions of the velocity field around an isolated building. For more complex geometries, the proposed framework performed well in regions where the flow properties were covered by the training dataset. Moreover, the present framework provided a continuous and smooth velocity field distribution in highly complicated applications, underscoring the robustness of the proposed turbulence modeling framework.