Purpose: In the γ evaluation algorithms reported so far, the exhaustive search for minimum distance over a certain maximum radius in the spatial‐dose space is required. The purpose of this work is to propose an algorithm based on the kd‐tree for nearest neighbor searching instead of using the sorted list to improve computational time for the γ evaluation of multi‐dimensional dose distributions. Method and Materials: The adaptive mesh data structure in our previous work was modified into the kd‐tree, which is essentially a binary tree structure. A kd‐tree was built from the evaluation distribution in which every leaf node is a k‐dimensional point in the spatial‐dose space, and every non‐leaf node has a splitting hyperplane that recursively divides the space into two subspaces. More spatial‐dose points were inserted into the kd‐tree based on the local dose gradient in the evaluation distribution by interpolation. To compute the gamma index, the nearest Euclidean distance was searched globally in the kd‐tree constructed from the evaluation distribution for each position‐dose point in the reference distribution. Results: Simulated 2D and 3D dose distributions similar to which described in Low et al. were used to evaluate the performance of our algorithm. We found that the building time for a kd‐tree is proportional to O(logN), where N is the pixel number of the evaluation distribution. The nearest neighbor searching time, i.e., the gamma calculation time, is proportional to O(N 1/k ), comparing to the exhaustive searching time (O(N)) based on a sorted list as demonstrated in this work and other's. Conclusions: The γ evaluation algorithm based on the kd‐tree nearest searching is more efficient than the sorted list based exhaustive searching. The algorithm could be used in the one‐time preprocess for all nearby simplexes in the geometric method proposed by Ju et al..