Multivariate time series (MTS) data sets are common in many multimedia, medical, process industry and financial applications such as gesture recognition, video sequence matching, EEG/ECG data analysis or prediction of abnormal situation or trend of stock price. Multivariate time series clustering is an important task in time series data mining. The unique structure of time series makes many traditional clustering methods unable to apply directly. In order to efficiently perform clustering for financial MTS datasets, we present a clustering approach based on locally linear embedding (LLE), which first converts the raw time series data into lower dimension by using LLE algorithm, and then applies a modified kmeans algorithm to the extracted feature vectors. Several experiments on a financial MTS database are performed and the results show the effectiveness of the presented method.