The forecasting of financial volatility is important for asset management and portfolio selection. It is difficult to forecast financial volatility accurately due to the high nonlinearity and clustering in financial volatility sequences. To improve the forecasting accuracy for financial volatility, Least squares wavelet support vector regression machines (LS-WSVR) is applied to forecasting financial volatility. Using daily SZSE fund index data from China stock markets and selecting three different kinds of wavelet kernel functions, the paper demonstrates the validity of LS-WSVR for fund volatility forecasting. Four statistical indices, RMSE, MAE, LL, and LINEX, are adopted to test the forecasting performance of LS-WSVR. Experimental results show that on the whole LS-WSVR with three different wavelet kernel functions outperforms the LS-SVR with Gaussian kernel function for in-sample and out-of-sample fund volatility forecasting. Moreover, the forecasting performance of LS-WSVR with Morlet kernel function is better than those of LS-WSVR with Mexican hat wavelet kernel function and DOG wavelet kernel function