Torrefaction of biomass is a thermal treatment process executed in an oxygen poor environment in order to improve storage and combustion properties. This paper reports the results of multivariate analysis of data obtained from a continuous torrefaction trial conducted in an integrated pilot plant (100 kg/h capacity) comprising automated feeding, drying, torrefaction, torrefaction gas combustion and product cooling. The purpose of the work was to gain insights into the process behaviour that might be useful in the conception of a process controller that makes it possible to obtain a product having a constant calorific value from heterogeneous raw biomass having a fluctuating calorific value. The questions addressed in this paper are; what are the most important predictors of final product calorific value, what parameters should be manipulated, what is the lag time between parameter adjustment and response, which manipulated parameters account for deviations from mean process values during operation and what is the predicted product calorific value as a function of the important manipulated parameters? Time series data collected during 35 hours of continuous operation and composed of observations of 95 process variables including 23 on-line temperature measurements and 34 off-line measurements of raw and torrefied biomass properties such as moisture content and calorific value were analysed. The results of PCA and PLS analysis identified the dryer temperature as having the most important influence on process stability and the temperature of the central torrefactor section as having the most important influence on product calorific value.