The increase in credit card transactions has inevitably caused an increase in credit card fraud. A total of 157,688 fraud cases occurred in 2018 worldwide, causing a total loss of $24.26 billion. This paper proposes using two types of autoencoder models to detect credit card fraud. The first type uses reconstruction error to detect anomalies in the data. The model detects fraud by defining a threshold in the reconstruction error to flag the transactions as legitimate or fraud. The second type performs dimensionality reduction to encode the data and removes noises. The encoded data were then used to train three other models: K-nearest neighbours (KNN), logistic regression (LR), and support vector machine (SVM). We then applied these models to a European bank's imbalanced credit card data set. A comparison was made between the two autoencoder types and three baseline models: KNN, LR and SVM. The results showed that both autoencoders gave a good and comparable performance in detecting credit card frauds.