The study investigates the challenge of accurately interpreting electroencephalogram (EEG) signals generated during motor imagery tasks, which are central to brain–computer interface (BCI) research. To address this, the authors present a novel framework, termed 3D-CLMI, that integrates three-dimensional convolutional neural networks (3D-CNNs) with Long Short-Term Memory (LSTM) units enhanced by attention mechanisms. The 3D-CNN component is responsible for identifying spatial dependencies across EEG electrode arrays, enabling the extraction of localized and global topographical features. In parallel, the LSTM network is designed to capture the sequential and dynamic aspects of the signal, with the attention layer selectively emphasizing the most informative temporal segments. By merging the outputs of these complementary pathways, the framework generates a joint feature representation that improves classification performance. Empirical testing on the benchmark BCI Competition IV-2a dataset demonstrated a classification accuracy of 92.7%, which surpasses the performance of many existing state-of-the-art methods. Additional validation on data collected from a separate group of twelve participants confirmed the model’s ability to generalize across datasets. These findings underscore the value of combining spatial and temporal analyses for improving the robustness of motor imagery EEG classification.