A clustering algorithm based on the commonly used least-squared Euclidean distance performs poorly on AVHRR images of clouds. This is primarily due to the data inadequately fitting the assumptions on which such an algorithm is based. Algorithms based on two modified clustering criteria are shown to be convergent within the same algorithm shell. These modified criteria have been developed elsewhere to allow generalized Gaussian clusters and also to account for differences in the populations of the different clusters. The new algorithms are tested on satellite data and found to give much improved results.