Abstract: As an important aspect of spatial relation similarity, directional similarity is widely used in pattern recognition, spatial query, spatial data match, quality assess of cartographic generalization and consistency checking of multi-resolution spatial data. Therefore, calculation of directional similarity is necessary. Now,the influential calculating model of directional similarity is proposed by Goyal, which based on the direction-relation matrixand derive the similarity value by calculating the least cost for transforming one direction-relation matrix into another. However, the current model is complex and the calculating result is inconsistent with the recognition of people.To overcome the limitations of the model given by Goyal, an improved model for calculating the similarity of spatial direction relations between areal objects is proposed. The new model improves the model given by Goyal in three ways: Firstly, the direction-relation matrix model for calculating the spatial direction relations between simple areal objects is extended to accommodate the calculation of the spatial direction relations between areal object groups. Secondly, the definition of direction relation distance between single-element direction-relation matrices is improved according to cognitive knowledge of direction relation difference. Finally, a new method based on the minimum element to calculate the distance between two multi-element direction-relation matrices is presented, which simplifies the process of calculation. Two experiments are performed to validate the effectiveness of the proposed model. One experiment is recognition experiment by comparing the calculating results with the recognition results. The second experiment is an application experiment of the improved model. Experiments illustrate that the similarity result between two spatial direction relations calculated by the proposed model is consistent with the human recognition of directional relation differences. It is simple and feasible for applications.