Spatial association rule is one of the upmost knowledge rules in the result of spatial data mining. It emphasizes particularly on confirming the relation of data in different fields. It tries to find out the dependence of data in multi-fields. As we know, in GIS the spatial database is often separated into several layers or tables according the type of the spatial object such as road layer, building layer, plant layer etc. In the relational database we often separate it into several tables which be associated by the primary key and foreign key according the normal form theory. Consequently, the spatial data is stored in different layers and tables. It is necessary and meaning to mining the knowledge and rules in multi-layer and multi-tables. And, It is inevitable to mining spatial association rules in multi-layer in some application. There is a problem in it, that is the number of the rules are magnitude. So, we point a new way by using the cell pattern of the rules which the user interested to reduce and simplify the operation. In this paper the concept of multi-layer spatial association rule is put forward. Then an algorithm of mining multi-layer spatial association rule is presented which based on cell pattern and spatial concept relation. It was called AP-MLSAM in the paper. Last, an example in GIS is given. In AP-MLSAM, First, it confirms the patterns and rules that the user is interested in. Second it counts the large itemsets according with the cell pattern in each data layer. Last, the spatial association rules are gained by the itemsets which be counted in the second step. From the experiment, it proved that AP-MLSAM is effective. It improved the efficiency by reducing the time of finding the large itemsets. It is a significance research field for mining multi-layer spatial association rules. There are many applications based on multi-layer spatial association analyse. For example: traffic flux analyse in city, weather pattern analyse, trend analyse for climate and plant. All these applications request mining the association rules in the mass data. It is necessary to improve the efficiency of the algorithm. And this paper offers a new way to mine multi-layer spatial association rule based on concept relation using cell pattern.