The increasing prevalence of single input multiple output systems has created increased interest in equalization across multiple channels. An important step to creating an appropriate equalizer is to have a good estimate of the effective channel order. Traditionally, the selection criteria have used eigenvalues or singular values for estimation purposes. Using these values ignores important information that can be obtained from the nullspace of the least squares method's matrix. The nullspace has a structured basis that can be utilized to quantify the quality of channel coefficient estimates. These residual error measures cluster in the nullspace and spread in the signal space, when compared to the singular values. Those properties allow the residual errors to be substituted into information theoretic criteria for improved performance. Additionally, a simple exponential curve fit is used to further improve the estimate of the effective channel order. The new criterion outperforms other known criteria in a wide variety of scenarios at both high and low SNR.