There are some methods for searching in large music databases, like searching by song name or artist name. However, in some cases these methods are not enough. For instance, a person might not remember the name of a song, but might remember its melody. In Music Information Retrieval, there is a task called Query-By-Humming, which allows retrieving a rank of songs that are similar to an audio humming. In this research, we propose the use of vocal isolation methods to improve query-by-humming systems. To achieve this, different configurations of Query-by-Humming systems were tested to analyze the results and determine in which cases our proposal works better. The results showed that vocal isolation improves the performance of Query-by-Humming systems when the music collection consists of modern songs.