Knowledge of neural receptive fields helps decode animal's movement from neural activity in motor brain machine interfaces (BMI). The traditional tuning depth is a coarse metric to evaluate the neuron's tuning properties. The major difficulty of the tuning depth is that it does not allow for a direct comparison among neurons that have very different firing rates. An information theoretic tuning depth based on mutual information is proposed to evaluate how neural spikes encode the kinematic direction. This metric is scale invariant and makes the tuning analysis comparable among not only the cortical areas but also among kinematics vectors. It can help identify candidate neuron subsets to reduce the complexity of analysis and decoding in BMI. The application of the metric on neural encoding modeling also provides a way to estimate time delays in motor cortical neural activity.