Recent progress in Brain Machine Interface (BMI) research using signals from subdural electrocorticographical (ECoG) arrays implanted over primary motor cortex has demonstrated the potential for successful decoding of participant’s arm or hand movement. Concurrent advances in wireless implantable electrode technology are further shortening the gap between existing systems and real functioning rehabilitative neural prosthetics for use by patients with motor system disorders. However the long-term stability and effectiveness of BMI technologies depends crucially on the stability and effectiveness of the feature extraction and decoding algorithms used to translate the neural signals into behavioral commands. Given the limited density and coverage of ECoG channel arrays, it is critical to reliably extract as much of the available information as possible. While some success has been achieved using channel-based features, we have recently shown that improvement may be gained by first applying the Independent Component Analysis (ICA) method to the ECoG channel data to separate and isolate relevant cortical signals. In this talk, we will present some preliminary results of application of ICA to ECoG data for the derivation of features for BMI applications and comparison to standard approaches. We discuss the interpretation of the ICA features and suggest directions for future improvements.