A new channel order estimation method is proposed for single-input multi-output (SIMO) systems. The method is based on a cost function which is constructed from the channel output error (COE). Proposed method uses the least-squares-smoothing (LSS) technique for channel estimation and Moore-Penrose pseudoinverse for the estimation of input. Channel outputs are obtained using the estimated channel coefficients and the input data sequence extracted from the estimated input data matrix via data unstacking. The cost function is calculated from the difference between the observed and the estimated channel output. It is proven that the cost function has a global minimum. The proposed method performs significantly better than the alternative algorithms. The proposed algorithm is more robust to changes in channel order and the number of channels, and gives exact result from limited number of samples in case of free noise.