This work studied the feasibility of near-infrared spectroscopy (NIRS) with a fiber-optic probe for the prediction of state variables in solid-state fermentation (SSF) samples with neither previous treatment nor manipulation. The models were developed using 50 samples and had the following root mean square error of cross-validation (RMSECV): 0.0253 for moisture content, 4.720 mg/g wet medium for biomass, 3.51 FPA/g wet medium for cellulase. Ten external samples were used to test the stability of the prediction models. The maximum absolute error and maximum relative error (%) were 0.0372 and 5.86 for moisture content, -6.51 and 11.63 for biomass, and 2.26 and 8.52 for cellulase. The predicted values using the NIRS technology were similar to the values obtained using chemical methods. (C) 2008 Elsevier Ltd. All rights reserved.