Process model discovery algorithms are capable of extracting process models from event logs, but different algorithms have varying capabilities in handling event logs. Currently, most research on evaluating these algorithms involves indirect evaluation methods, which have limitations. To address this issue, a method is proposed to directly evaluate the reliability of process model discovery algorithms, using reliability as an important evaluation metric. The original event log is preprocessed to obtain an incremental sub-log collection, the process model discovery algorithm is applied to the incremental sub-logs and the original event log to obtain process models, and the reliability of the business process model discovery algorithm is evaluated through quality assessment. Based on nine public simulation event logs and four real event logs, multiple model discovery algorithms are experimented on from the aspects of weak reliability, noise interference reliability, and strong reliability. The experimental results show that the reliability values of Heuristic Miner, Inductive Miner-infrequent, Inductive Miner, and Alpha Miner are 4, 3.2, 2.4, and 1.6, respectively. Higher reliability values indicate stronger reliability of the algorithms. Thus, the proposed method can effectively evaluate the reliability of the algorithms.