Abstract. Go is a board game that originated in China thousands of years ago. It is considered a grand challenge problem in artificial intelligence, as top Go programs are currently unable to win against even highly skilled amateur human players. Top Go programs use the Monte Carlo Tree Search algorithm in combination with machine learning and domain-specific knowledge to select their moves. This research focuses on efficient storage of local configurations of stones, called patterns. Skilled human players use patterns in deciding where to play. For computers, storing all patterns is impossible for larger sizes due to memory and search speed limitations. However, only a small subset of all possible patterns are legal, and, of those, only a subset are useful. We present a technique for quickly and efficiently storing and retrieving these useful patterns.