This paper develops a robust method to estimate channel order in a single input multiple output (SIMO) or an oversampled single input single output (SISO) system. The method is based on the application of linear prediction filter theory. Current methods based on the use of information theoretic criteria suffer from estimation errors at high SNR or when sub-channels are correlated. The new algorithm exploits (spatial) oversampling and the fact that the linear prediction approach to channel estimation is robust to the over-determination of the channel order. The method developed is robust over a wide SNR range and is independent of correlation in the received channels. Simulation results are presented showing performance and behaviour at various signal to noise ratios.