This study summarized the application field of machine learning to explore the basic seismic thinking of machine learning methods used for bridges. The development and actuality of bridge seismic analysis theories and technology were briefly reviewed, particularly in relation to the field of civil engineering. This study introduced the concept of machine learning, summarized its key factors and current software platforms, and illustrated the common methods and representative algorithms of machine learning using simple examples. First, the normal types of simply supported and continuous girder bridges, which are used for highway bridges with small and medium spans in China, were summarized. The earthquake damage phenomena had been observed for bridge types during the Wenchuan earthquake. Data from these phenomena were assessed, including the damage grade divisions of piers, bearings, shear keys, and abutments. Second, a series of seismic performance tests had been conducted by international and domestic academics for bearings, shear keys, piers, and abutments. These tests were summarized in this study to obtain the constitutive relationships for seismic analysis and determine the seismic design parameters of bridge components (including foundations). Finally, an overall analysis methodology based on machine learning was introduced into the bridge seismic analysis. This methodology explained that machine learning for bridge seismic tasks had two aspects. The first was the collection of considerable bridge design data and set up data sets, and the second was data reduction, including raw data processing and debugging or developing a reasonable machine learning algorithm model. This study also discussed the shortcomings of existing performance-based probabilistic seismic design and evaluation methods that are currently used in analyzing bridges in China. Results indicated the potential future major concerns for bridge seismic analysis technology. The establishment of computational simulations based on artificial intelligence method was also recommended. In addition, the mutual integration between disciplines and mutual communication among different professionals were advocated in this study.