ERROR: Initial covariance is required, even though I am giving it the covariance in gmdistribution.fit.

I am trying to fit a Gaussian mixture model for data, I am calculated inital means, covariance and Pcomponents , and created a structure S. Then I use this command to fit the model.

    opts = statset('MaxIter', 30);
    obj = gmdistribution.fit(data,maxk,'Start',S,'CovType','diagonal','Options',opts);

Where S is a structure which has the means, covariance and Pcomponents in the desired format. But when I run it i always get the error.

Error using gmcluster>checkInitParam (line 384) The initial covariance is required.

Error in gmcluster (line 135) initPara=checkInitParam(start,k,d,CovType, SharedCov);

Error in gmdistribution.fit (line 174) [S,NlogL,optimInfo] =...

But I am giving it initial covariance, any input on this will be appreciated.

------- Eric