We propose a likelihood test for a covariance estimated from sample data which is used to determine the number of narrow band source signals. This Minimum Description Length (MDL) estimator is shown to be robust against deviations from the assumption of equal noise level across the array. A number of source Direction-Of-Arrivals (DOA) greater than the number of physical array elements is of interest. We propose a novel spatial smoothing algorithm which accurately estimates parameters for the covariance likelihood test. Improved parameter estimation performance is achieved when compared with the conventional spatial smoothing algorithm. Finally, the proposed source number estimator is validated through numerical results and compared with other recently developed approaches.