Recommender systems provide personalized recommendations to users and are widely used commercially in e-commerce and streaming services, when the num- ber of available products can be overwhelming for the users. In the domain of board games, sites such as BoardGameGeek provide information about thou- sands of games, but without personalized recommendations. Recommender sys- tems for board games have been studied in a few studies, but further research is needed to fully leverage the benefits of recommender systems in this domain. In this study we explored how the number of neighbors (value of k) in the k- nearest neighbor, a machine learning algorithm commonly used in recommender systems, affect the accuracy in a recommender system for board games. We im- plemented a recommender system based on collaborative filtering, using data from BoardGameGeek with user ratings of board games. To evaluate the sys- tem, we performed experiments where we measured Mean Absolute Error, Mean Average Precision and Mean Average Recall. We got mixed results in the study, partly due to lack of standard for evaluation. Mean Absolute Error was high and marginally improved with increased value of k. Mean Average Precision was either low and decreased when the value of k increased, or it was high and marginally increased. Overall the results of the study indicated that the accuracy of the recommender system was high and marginally increased with increased k in the k-nearest neighbors algorithm.