The automatic diagnosis for sucker rod pump (SRP) is an essential measurement to ensure the oil fields' interests in the oil recovery. As the important information resource on monitoring and diagnosis, the dynamometer card (DC) plays an irreplaceable role in oil engineering. In the application, how to use DC to fulfill the diagnosis is always the key to this problem. Thus, a novel method based on load analysis and Bayesian network is proposed in this paper. At first off, DC's coordinate is transformed to cater to the load analysis, which provides an instinctive way for analyzing. After that, five statistical features and Shannon entropy are extracted from the DC, which are employed as the input of the Bayesian network (BN) presented in the particular framework. At last, a set of field dynamometer card is employed as the experimental data and the experimental results demonstrate the feasibility and superiority of the proposed method for diagnosing the working states of SRPs.