Condition evaluation of relay protection has always been the focus and difficulty of condition based maintenance. This paper proposes a condition evaluation method for protection based on fuzzy support vector machine (FSVM). First, via acquiring historical data of relay protection devices in smart substations, condition evaluation factors are chosen appropriately which will be pretreated next, and training set is generated by calculating the remaining life in percentage according to the inspection and failure record. Then regression training of FSVM is conducted to generate a model for online condition evaluation by using online inspecting record, and the worst index will be used to represent the real-time state of the device. Simulations show that, by introducing the sample weights, the method can distinguish the difference between devices which makes the algorithm take both the device''s personality and family generality into account, and have certain anti-noise ability at the same time.