A new method named multi-reference-state updating (MRSU) pertaining to the dynamical analogue prediction, is developed on the basis of previous studies on analogue-dynamical models, in order to further effectively utilize the available information of historical observation data. In this scheme, according to a new idea of “updating", it is required that multi-reference states are renewedly selected on the period of analogue updating in the process of the analogue-dynamical model integration, and optimal forecast vectors are estimated from multi-forecasts produced by analogue-dynamical model by employing the hyperplane approximation method. Such “selection-estimation" cycles are repeatedly operated until the whole forecasts are completed. Furthermore, the simplified MRSU is applied to the T63 global spectral model, and the results of monthly forecast experiments show that for the daily and monthly mean circulation, the MRSU can effectively reduce forecast errors and improve forecast skill compared with the control forecasts.