This paper evaluates the performance of a source detection procedure that uses Akaike's minimum Final Prediction Error criterion. The procedure selects the order of a state-space model of the array signal, which, in turn, forms the basis of H_infin optimal single-snapshot direction finding. The authors analyze source detection as a sequential binary hypothesis test and estimate its effective "false alarm rate." The paper also presents detection results from synthetic data generated by the Sonar Simulation Toolset, which corroborate the authors' findings.