The basis for the determination of diabetes mellitus is the classification studies that constitute the infrastructure of clinical decision support systems. The main purpose of classification studies is to increase the classification performance and increase the diagnostic rate. Different classification methods and different optimization algorithms are used for this. In this context, in this study, a classification study with Autoencoder deep neural networks was performed for the diagnosis of diabetes mellitus. The Pima Indian diabetes dataset in the UCI machine learning laboratory, which is widely used in the classification study, was used. The results of the study were compared with the results of previous which focuses on the diagnosis of diabetes studies using the same UCI machine learning dataset. The obtained classification accuracy is 97.3% and higher than the previously mentioned classification methods. The obtained evaluations show that the proposed method is very efficient and increases the classification success.