This means that the population model that you specified implies a non-positive definite (improper) covariance matrix. For example, you may have specified your population model in such a way that two or more observed variables are perfectly correlated in the implied correlation structure. Or there may be linear dependencies among observed variables such that one or more variables are perfectly linearly dependent on other variables. You need to change the parameters of your population model so that the implied matrix becomes positive definite. This can be tricky sometimes because the issues with the model-implied covariance matrix may not be obvious.