We search for various things every day - food, information on the Internet or someone's name in memory. Despite the different nature of these tasks, they all have a common feature - a final goal with an unknown location in a complex environment. This property of the search raises a problem of trade-off between exploration of new opportunities and exploitation of the known information. We used the data from the semantic fluency task experiment to investigate how humans switch between exploration and exploitation strategies when they search in memory. On comparing different search models, the one that assumes that humans switch strategies according to the semantic quality of the current neighbourhood best fits the data. Moreover, participants who set higher thresholds for the words with better quality of the neighbourhood tend to retrieve more words. We also used regression analysis to find out which factors affect efficiency of both strategies.