Abstract: Abstract:The paper presents a new way to capture the multi-peaks of financial dates, which is the nonparametric kernel density estimation with weishted samples. The new way captures the characters of multi-peaks and clusters of financial dates to be facility than the ARCH class and mixture distributions models. For example, we find evidences for the multi-peaks of the dynamical structures of stock index in shanghai stock market and some individual stocks. Monte Carlo simulations indicate the weights are important to describe returns which are mixture distributions