It is of great interest to evaluate the ability of the current generation of climate models to simulate observed extreme rainfall distributions and their trends. But the lack of comparable long-term global gridded daily observations often leads to a deferral of model evaluation or limited evaluation of only the mean precipitation climatology. Additional difficulties arise from the scaling issue when comparing the extreme events among the observation and models with different spatial resolution. The interpretation of model output as point estimate vs. areal mean and the spatial interpolation schemes both can have strong impact on the outcome of validation and comparison. With appropriate consideration of the spatial scale of the observed and simulated data, the present-climate extreme precipitation events simulated by IPCC AR4 climate models are evaluated. For the continental US and East Asia, the daily extreme rainfall distribution is reasonably simulated. The common model bias is the reduction of spatial variability (underestimate in higher extremes and overestimate in area where daily extreme rainfall is small). The bias resemble to the systematic error in the annual mean precipitation simulation. The model simulations for the past trends in the extreme precipitation over the 1961-2000 period are inconsistent with observation (in US and East Asia) and no common features found among different models. Model ensemble mean projections for the 2081-2100 period show that, except subtropical arid region, extreme daily rainfall almost increase everywhere with larger percentage increase in the tropics. With the model reliability information, the common characteristics of future projection of simulated changes in extreme precipitation from IPCC AR4 models and their uncertainties will be assessed.