The BNCI 2014-002 Motor Imagery dataset comprises EEG recordings from 14 healthy subjects performing two-class motor imagery tasks (right hand and feet imagination) in a cue-guided Graz-BCI paradigm. Data were acquired at 512 Hz using 15 EEG channels with online Butterworth filtering and Laplacian montage, yielding 160 trials per subject with continuous visual feedback. This minimally preprocessed dataset has been benchmarked with multiple machine learning classifiers (Random Forest, Shrinkage LDA) achieving median accuracy of 80.42%, and serves as a standard evaluation resource for brain-computer interface research.