GMM gives different results for the same file

Hi everyone, I construct a Gaussian Mixture Model with a dataset by the command: model = gmdistribution.fit(data,M,'Replicates',5);

I test new data with the command: [P, l] = posterior(model,new_data);

The problem is that I don't have the same result from the system. I use the log-likelihood to make decision, and this value for the same data (new_data) in different run is differ. What does it dependo on? How can I resolv this problem? Thanks a lot.