Volatile compounds released from herring fillets (Clupea harengus) during 15 days of storage on ice have been measured with a commercial hybrid gas-sensor array system. Using partial least-squares regression modeling, the sensor responses were correlated with data from chemical analyses (lipid oxidation products and antioxidants) and sensory analyses (odor). Eight of the 16 sensors proved significant in the correlation studies: 6 metal oxide semiconductor field effect transistor (MOSFET) sensors and 2 Taguchi type sensors. Correlation coefficients for chemical and sensory data ranged from 0.9 to 0.98 and from 0.49 to 0.92, respectively, with 0.92 referring to both "sharp/acrid" and "rancid" odors. Prediction errors ranged from 8 to 14% and from 11 to 25% for the chemical and sensory measures, respectively. That the prediction errors for oxidation product formation (5-9%) were close to the analytical errors of the chemical reference methods indicated close to "optimum" performance of the gas-sensor system. The sensor system predicted the storage time of the herring with a 1-day error. Results illustrate high potential of the gas-sensor technology in rapid nondestructive quality determination of ice-stored herring.