This paper explores the use of convolutional neural networks architectures in the context of volcano seismic events classification through the use of grayscale spectrogram images of long-period and volcano-tectonic seismic events. We combined three different architectures with a set of hyper-parameter configurations that produced 720 classification models, which were able to learn the morphological pattern described by the grayscale spectrogram images of seismic events. Downscaling of all grayscale spectrogram images was used to reduce the computation time for each model without losing performance and avoiding any overfitting. The explored architectures provided good results in terms of the area under the receiver operating characteristic curve scores. However, when considering the accuracy scores in the selection process, the best model to classify grayscale spectrogram images of both types of seismic events was the CNN3 architecture with a (3 × 3) convolutional and pool kernel size, same padding type and 1×10-4 as learning rate, which achieved an area under the receiver operating characteristic curve and accuracy values of 0.94 and 94.13%, respectively.