At present，most industrial online coal analyzers utilize intense energy beam like nuclear radiation scanning，the test utilities are complex，the use and maintenance cost is high，so the on-line detection of incoming coal quality is generally not realized in the field of thermal power in China. Due to the coal quality fluctuation control inefficiency of power generation process caused by coal quality fluctuation，and soft measurement technology for coal quality monitoring begins attracting attention. Utilizing distributed-sensors in the DCS system，soft-measurement of fuel quality of coal fed into the boiler are realized through mechanism analysis and data learning，which meets application requirements for coal-fired power generation with few additional investigation cost. However，coal quality has multiple indexes，iden tification of different quality index applies different approaches，the soft sensing of incoming coal is lack of systematic technical system demonstration. In this study，the classification and fundamental basis of soft-measurement for coal quality parameters were summarized，and the error analysis of online soft sensing of coal quality was analysed. At last，suggestions of future technological developments of soft-measurement of coal quality were proposed. According to methodological character，soft measurement technologies for coal quality are divided into two types，one type based on mechanism analysis and the other based on machine learning. Soft-measurement of coal quality is more fitted for power generation application. The real-time tracking of the actual coal quality can realize effective regulation of thermal power production process. Coal quality is constructively multiple system，soft-measurement of coal quality based on mechanism analysis builds fundamental models through process analysis of pulverizing，burning，heat and energy transfer，different technical trends are formed. For such technology，explicit models are offered，but the analysis process is easily effected by multiple factors，it is hard to reach accurate results especially for frequent load regulation condition. Coal quality measurement based on machine learning avoids complex mechanism analysis，while on the other hand，sample data set preparing and intelligent modelling facing high requirements for goal of efficient application.