In this paper we present WoMine-i, a novel algorithm to detect infrequent behavioural patterns from a process model, measuring their frequency with the instances of the log. A behavioural pattern is a subgraph of the process model, com- posed by all type of structures —sequences, selections, parallels and/or loops—, which represents the behaviour of a part of the process. And it is considered infrequent when its complete execution happens in a number of cases from the log below a predefined threshold. To find the infrequent patterns, WoMine-i performs an a priori search starting with the minimal structures of the model. In this search, there is an expansion stage done in two ways: i) adding other minimal structures not contained in the current pattern, and ii) adding arcs. This expansion is followed by a pruning strategy that verifies the upward-closure property of support —also known as monotonicity. This property ensures that if a pattern is infrequent, all patterns containing it will be infrequent and, thus, it is no necessary to continue expanding it —the minimum pattern itself expresses all the infrequent behaviour containing it. This pruning presents an exception in order to simplify the results: If a pattern is infrequent and maintains the value of its frequency with the expansion, it is not removed from the expansion stage —it means it is being expanded with a selection branch with less frequency. In this way, WoMine-i returns the largest patterns expressing the minimum infrequent behaviour. In each step of the iterative process, WoMine-i reduces the search space by pruning some of the generated patterns. For this, an algorithm to check the frequency of a pattern is needed. WoMine-i generates the different paths of a pattern, henceforth simple patterns, and checks the frequency of each one. To measure the frequency of a simple pattern, WoMine-i replays the traces of the log and checks how many of them are compliant with it. At each step of the replay, the algorithm checks if the simple pattern is being correctly executed. Afterwards, WoMine-i assigns to the pattern the higher of its simple patterns frequency, and checks if the pattern is considered frequent w.r.t. the threshold. Thereby, given a log and a corresponding model, WoMine-i is able to return the infrequent patterns of it. Allowing to study them and enhance the process consequently.