Cytogeneticists can diagnose various health and genetic disorders based on chromosome analysis. The standard technique in chromosome analysis is karyotyping, where each chromosome is classified into one of 24 chromosome classes. This process is performed manually inside the cytogenetics laboratory, and it consumes time, effort, and needs domain expertise. We automate in this paper a chromosome classification task by fine-tuning pre-trained convolutional neural networks models (VGG19, ResNet50, and MobileNetv2) and ensemble their results using majority voting and average voting. We compare the empirical performance for both ensemble methods on the biomedical imaging laboratory dataset that contains 5474 chromosome images which are publicly available online and on the diagnostic genomic medicine unit dataset that contains 6011 chromosome images. The best classification accuracy obtained on the biomedical imaging laboratory, and the diagnostic genomic medicine unit datasets was 97.01, 94.97%, respectively, when ensemble the results of the models by average voting.