The estimation performance of a soft-sensor would deteriorate when the process characteristics change.To cope with such changes，recursive PLS and just-in-time modeling have been developed. However，these methods update a soft-sensor model whenever the new input-output data are sampled no matter whether the estimation performance of the soft-sensor deteriorates or not.Therefore，an approach for adaptive correction of a soft-sensor was proposed based on model performance monitoring and assessment.Firstly，an estimation performance assessment index of soft sensors was designed to objectively detect the degradation of a soft-sensor relative to their design performance.Then，the statistical limits of the index could be determined based on the time-series index data collected under normal conditions. The soft sensor would be updated only when the assessment index exceeded the corresponding statistical limits.A state classifier could diagnose the type of change of process characteristics according to the discrete Fourier transforms of the indexes series.The adaptive correction method would update model recursively when process characteristics were gradually changing and reconstructed the local model in a neighborhood around the query point when process characteristics had an abrupt change.The usefulness of the proposed method was demonstrated through a case study of a continuous stirred tank reactor (CSTR)process and the real industrial data of an industrial evaporator from DAISY (database for identification of systems). The result showed that the method could avoid large fluctuation of measurement caused by blind correction and serious off-line effects of analysis noise.The method could cope with changes in process characteristics and improve the estimation performance，and could have the potential for realizing efficient maintenance of soft-sensors in the actual industrial process.