In this paper, we found that the distribution of residuals (i.e., the difference between the estimated value of state estimation model and the sensor reading) of a sensor in abnormal state would change significantly comparing with the sensor in normal state. So, we proposed a health assessment technology based on the dissimilarity of residual distributions. The distance between these two distributions can be used as the health indicator of the sensor. The greater the distance, which means the less similar the two distributions are, the worse the health status of the instrument. The calculation results showed that the dissimilarity of residual distributions had certain universality as a health indicator of sensor, and it could accurately reflect the health state of a sensor. This technology can judge the degradation of the sensor in advance through the real-time online health assessment. By using this technology, the safety of nuclear power plant can be improved and the operation & maintenance cost of nuclear power plant can be reduced. In the later stage, the combined distribution of signal residuals from multiple sensors can also be used to evaluate the health of a target system.