The evaluation of quality grade of groundwater involving various uncertainty factors is of fuzzy, random and discrete characteristics. In order to reflect distribution characteristics of evaluation indexes and improve the rationality and reliability of groundwater quality classification, a connection cloud model coupled with extension theory was proposed here to describe conversion situation in the classification ranks. Firstly, digital characteristics of connection cloud model based on the classification standard were identified, and the connection cloud mapping in the finite intervals was generated to simulate the classification standard. Namely, certainty and uncertainty relationships between the measured evaluation indicators and their quality grades might be depicted by a connection cloud from a unified perspective. Then combined with index weight, the extension matrix based on the connection cloud was constructed to analyze the relationship between measured evaluation indicators and quality grades. Next, quality grade was specified by the comprehensive cloud correlation degree, credible degrees of the evaluation results were also given. Finally, case studies and comparison with the projection pursuit based on fuzzy matter-element method were conducted, the results with less than 0.01confidence factor obtained by the model proposed here do well agreement with those by the projection pursuit method, and are more feasible and effective. Moreover, it can overcome the shortcomings of the extension theory that cannot reflect the fuzzy characteristic of the evaluation index.