There has been substantial effort dedicated to the issue of how to incorporate spatial information to improve the classification accuracy in past decades and some excellent methods have been developed. Each method has its own advantages and disadvantages for different images and user requirements. This paper proposes a new classification method, which introduces multiple-point simulation to improve the classification of remotely sensed imagery data by incorporating structural information through a training image. This new method named CCSSM is the derivation of two classifications and based on spectral and spatial information, which then are fused. For validation purpose, a real-life example of road extraction from Landsat TM is used to substantiate the conceptual arguments. An assessment of the accuracy of the proposed method compared with results using a maximum likelihood classifier shows the overall accuracy improves from 48.9% to 82.6%, and the kappa coefficient improves from 0.12 to 0.55 and therefore, the new method has superior overall performance on the classification of remotely sensed data.