Test cost and misclassification cost are commonly considered in costsensitive learning. In real applications, the test cost of a feature is often related to the granularity of attribute values, and the misclassification cost of an object with multiple attributes is usually influenced by the total test cost of attributes. Based on this consideration, the paper studies the selection of attribute and granularity of data in the case where the total test cost is restrained. Aiming at minimizing the average total cost of data processing, a method is proposed to choose the optimal attribute subset and the optimal data granularity simultaneously. We first construct the theoretical model of the proposed method, and then design an efficient algorithm. Experimental results show that the proposed algorithm can effectively select the attributes and the granularity of data under different test cost constraints.