A continuously-sized circuit resulting from transistor sizing consists of gates with a large variety of sizes. In the standard cell based design flow where every gate is implemented by a cell, a large number of different cells need to be prepared to implement an entire circuit. In this paper, we first provide a formal formulation of the performance-constrained different cell count minimization problem, and then propose an effective heuristic which iteratively minimizes the number of cells under performance constraints such as area, delay and power. Experimental results on the ISCAS 85 benchmark circuits implemented in a 90nm fabrication technology demonstrate that different cell counts are reduced by 74.3% on average while accepting a 1% delay degradation. Compared to circuits using a typical discretely-sized cell library, we also demonstrate that the proposed method can generate better circuits using the same number of cells.