Backscattering in coastal waters is stronger than that in open sea because of high suspended sediment，and its information is reflected well in remote sensing images. The key to estimate suspended sediment concentration（SSC） from remote sensing images is to establish the relationship between SSC and water reflectance R . Many theoretic or empirical models have been developed for calculating SSC. But most of them need real-time survey data of SSC，which limit the use of these models. In this paper，spectral mixing analysis method，which needs little real-time data，was introduced in detail. The pixels with minimun and maximal SSC were used as endmembers，and other pixels were regarded as the mixtures of the two endmembers. A linear spectral mixing analysis model was as followed： Rb =（1 - fhigh ）RLb + fhighRHb + Eb b = 1，2，?，B where Rb was the bth reflectance，RLb and RHb were the bth reflectance for minimum or maximal SSC，fhigh was the fraction of maximal SSC，Eb was residual error and B was the number of bands. In order to get the minimum sum of Eb （ b = 1，2，?，B），the following model has been developed： fhigh = Σ B b = 1 （Rb - RLb ）（RHb - RLb ） Σ B b = 1 （RHb - RLb ）2 Based on TM remote sensing images，the quantitative relationship between SSC and spectral reflectance was studied. The distribution of SSCs in coastal waters of Minjiang River was calculated by using spectral mixing analysis and empirical formula respectively. The results showed that：（1）the high SSC regions lay in the mouth of Minjiang River，and the further from the coastal，the lower was the SSC；（2）the distribution of SSC can be estimated quickly and objectly from remote sensing images；（3）spectral mixing analysis is a good method，especially in areas lack of much measured data，for it can make full use of multi-band of images and needs less samples.