Machine learning is a hot topic recently. For condition monitoring, the machine learning is mainly used to improve the failure diagnosis accuracy, as the machine learning can provide flexible decision function. This paper firstly discusses the advantage and disadvantage of the state of art condition monitoring methods. It figures out the machine learning techniques follow a general procedure and can be unified into a general framework. Later on, it figures out some key issues of applying machine learning to condition monitoring. Two examples are given to demonstrate the advantage and disadvantage of machine learning.