Multivariate volatility model play an important role in portfolio construction, asset pricing and risk management. In practice, since a large number of assets are considered simultaneously, the two most common models are the exponential weighted moving average （EWMA） model suggested in J.P. Morgan＇ s RiskmetricsTM and orthogonal GARCH （O-GARCH） model based on principal component analysis （PCA） of the return series. However, the assumptions used in both models are too restrictive. For instance, principal components （PCs） are unconditionally uncorrelated but not necessarily conditionally correlated, so their conditional covariance matrix may not be diagonal and O-GARCH model is not reliable in this sense. This paper puts forward a new multivariate volatility model, i.e., IC-GARCH model, based on the so-called independent component analysis （ICA）. It is expected that the conditional covariance matrix of ICs may look more like a diagonal one than that of PCs, which hopefully can remedy the defect of O-GARCH model. Two real data sets are used to illustrate the power of IC-GARCH model. The results from two mis-specification tests both demonstrate the advantage of IC-GARCH model over EWMA and O-GARCH models