This means that you have a negative residual variance for that variable. This in turn means either that you have extracted too many factors or that you have a small sample so that by sampling error you estimate a negative residual variance. Deleting the variable typically doesn't help because the problem may move to another variable.
In some applications you don't get a good model fit before negative residual variances start occuring. This implies that a factor model is not a sufficiently good approximation for the data at hand.