One of the main problems in chilean beekeeping is the late diseases diagnosis that affects beehives. In this work, convolutional neuronal networks are used to create a system that detect beehives health by classifying the sound they emit represented by spectrograms. A dataset is made from audio registers recorded in Chile. From this data, two models for beehives classification are elaborated with different architectures. The model implemented through Transfer Learning obtains a high percentage of accuracy (0.9303 in validation) at classifying recordings according to their health condition, which is comparable to other related publications about Machine Learning applied in beekeeping.