As an important branch of the spatial data mining, spatial co-location pattern mining refers to discovering a subset of the set of features whose instances are often neighbor in space. In many practical scenes, the instances of spatial features include not only location information, but also attribute information. Some scholars use the type-1 fuzzy membership function to mine fuzzy co-location patterns from spatial instances with attribute information. However, the type-1 membership function itself is uncertain. Therefore, there is some deviation in describing the membership degree of attributes of spatial instances by using a type-1 membership function. To solve this problem, we propose a fuzzy co-location pattern mining method based on type-2 fuzzy membership function. Firstly, we collected interval evaluation values of interval data of attribute information from 1000 experts, and formed granular data. Then, on the basis of the original type-1 membership function, a type-2 fuzzy membership function based on elliptic curve is expanded, and the parameters of the type-2 fuzzy membership function are adjusted by using a gradual method, so that the footprint of uncertainty (FOU) in the function meets the connectivity and the threshold given by the user. After that, we design a fuzzy co-location pattern mining algorithm incorporating type-2 fuzzy membership function into the traditional Join-based algorithm. In which, we define the concepts of fuzzy feature, fuzzy co-location pattern, upper bound participation index, lower bound participation index. In order to improve the efficiency of our method, we also put forward a pruning strategy. We have done a lot of experiments on synthetic and real data sets, which proves the effectiveness and efficiency of our proposed algorithm.