A novel approach for evaluating the tracking ability of photoelectric theodolite is proposed. Equivalent model of theodolite tracking error based on the BP neural network structure is identified. The Levenberg-Marquardt (LM) algorithm is adopted in the training method of BP neural network for the sake of speeding up training process. The equivalent sine signal is inputted to the model, and the output is gotten. The evaluation of tracking performance is obtained based on the statistical calculation of output. The estimate errors of equivalent model including average error, maximum error and standard error are 2.5872e-006°≈0°, 3.6″and 1.6″, respectively. The result shows that the equivalent identification model based on BP neural network meets the needs of evaluating the tracking performance of theodolite. The accurate evaluation of tracking performance is achieved.