In this paper, we test the superior predictive ability of Machine Learning models on the Bitcoin market, while adjusting the results for data snooping. Our findings show that Machine Learning models inspired from five different types of classifiers are capable of generating above-random predictions for future price changes and, consequently, positive excess returns over the na?ve buy and hold benchmark. However, when accounting for data snooping, the returns turn out to not be statistically and economically significant. Moreover, Machine Learning models underperform more simple and popular models derived from Technical Analysis, such as moving average cross-over rules. Even though the scope of the paper is limited, the evidence provided may be considered troublesome for investors considering applying ML-inspired trading models in cryptocurrency markets.