Abstract: With the development of information technology，mixed data should be processed which changes from the simple numerical or categorical to mixed value.Compared with clustering methods for pure numerical data，the methods can process mixed numerical effectively are very few.This paper presents a new algorithm based on lattice for multiple-attribute which uses lattice covers to calculate the similarity between two objects.In the algorithm，each sample will be distributed to the cluster where it gets the maximum cover count number.Experiment shows that the new algorithm is more efficient than the other classical algorithms.