In order to more effectively assess bearing performance degradation degree, a rolling bearing state evaluation method is proposed based on the similarity of the main manifold space curve. The high dimensional feature of vibration signal is extracted and then is converted to low dimensional space using the manifold learning algorithm of Laplacian eigenmaps (LE). Then samples from the curve according to soft-K principal curve algorithm are combined with the discrete Frechet distance to plot the condition assessment curve. Comparing with the hidden Markov model (HMM), deep belief network (DBN) method，small breakdown of equipment could be detected earlier，and the health state quantitative assessment of a rolling bearing is achieved.