In order to improve the forecast precision of daily water consumption,the paper proposed a forecast model based on multi-scale relevance vector machine (MSRVM). The non-stationary time series of daily water consumption are decomposed into different scales stationary time series by stationary wavelet transform.Then it established the regression model of relevance vector machine in each scale to predict respectively,and finally employed the forecast results of the RVM outputs at all the scales to reconstruct the forecast value of the original daily water consumption time series through the inverse wavelet transform. Application examples show that compared with mono-scale RVM model, the MSRVM model has the better precision and can be applied in the forecast of daily water consumption.