An algorithmic composition using genetic algorithms. The greatest difficulty in modeling genetic algorithms for musical composition is to create an fitness function capable of simulating human intentions in the evaluation of musical compositions generated. Therefore, in this work a fitness function was developed based on the similarity between the melodies generated and the melody of the algorithm, generating melodic variations from the original melody. The melodies created algorithmically were compared to the original melodies through ratings with people. The results show that the generated melodies remember the originals, however, adjustments are needed to better model musical patterns.