The influence of source number estimation on application of Independent Component Analysis (ICA) methods is researched.For positive model of ICA, when the estimated source number is less than the actual source number, the positive model is changed to over-determined model, and natural gradient algorithm is implied for simulation experiment.To the case that the estimated source number is more than the actual source number, the positive model is changed to under-determined model, for which the hybrid matrix estimation algorithm based on single source detection is used for estimating the hybrid matrix, and subspace representation algorithm is employed for studying the influence on source separation.Experimental results show that, as to the problem of positive model changing to over-determined model, in the case of signal to noise ratio satisfying the requirements, the number of successfully separated source signals is usually equal to that of estimated source number.As to positive model changing to under-determined model, usually all of the source signals can be successfully separated, but there is one or more of the separated signals is the copy of the original source signal.In this case, the separated signals can be merged based on correlation analysis, and all of the original source signals can be separated successfully.The results supply a referenced for applying ICA methods.