Statistical inference in factor analysis models is usually based on parametric representation and on maximum likelihood estimates. However , such approach is not always applicable. For example, the case when a sample covariance matrix is singular the likelihood ratio test cannot be used. One of the approaches for overcoming these difficulties consists in using model invariants, that is, polynomial equality relations that the model imposes on the entries of the covariance matrix of the observed variables. In this paper a simple method for finding model invariants for any number of latent variables is developed.