The available radiation time series employing the radiation time series method involves too few room structures and characteristic parameter numbers， which results in a large deviation of the design cooling load. Firstly， considering the commonly-used envelop structures and their characteristic parameters of buildings in China， the room samples combined with the structures and their characteristic parameters were sampled by a simple random method. Then， the heat balance method was used to calculate the radiation time series of each room sample. The main characteristic parameters that affect the radiation time series of a room are extracted by the CART algorithm， and the rooms were classified according to the main characteristic parameters. Finally， the K-Medoids algorithm is used to determine the representative radiation time series for each type of room. Using the proposed decision tree classification method， the non-solar radiation time series and the solar radiation time series of all rooms combined by the structures and their characteristic parameters were classified into twelve and eight categories， respectively. The applicability verification shows that the representative radiation time series for each category of rooms well represents all rooms in this category， which can significantly improve the accuracy of design cooling load.