The existence of feeding rhythms implies that fish would feed better during their preferred feeding phase but reject food at any other time. In the present paper, we tested the performance of a new device for continuously collecting and detecting uneaten food pellets. The device is basically made of two parts: a pellet collector placed just under the feeder and a decanter with a sensor attached to the bottom. When a food pellet was not eaten, it was rapidly collected and transferred to the decanter, where it was detected while dropping by an infrared sensor coupled to a microcomputer. To validate this system, five groups of fifteen sea bass, Dicentrachus labrax L., were maintained under natural conditions and subjected to a daily feeding cycle (feeding rate = 2.5% of body weight) consisting of three meals of one hour duration each (0800-0900 h, 1600-1700 h and 2400-0100 h). Uneaten pellets together with demand-feeding activity were simultaneously recorded. In addition to these test tanks, "natural" demand-feeding rhythms were also investigated in five groups of sea bass maintained under an ad lib self-feeding regime. In the test tanks, when submitted to the three meal feeding cycle, sea bass showed clear time preferences for feeding, since they fed mostly during the morning and afternoon, rejecting food at night. Consequently, the profile of uneaten pellets peaked at night butt remained very low during daytime. This diurnal preference for feeding is consistent with the almost strict diurnal feeding rhythm found in the sea bass groups under ad lib self-feeding. These results revealed the usefulness of this device in estimating food utilization and its potential application in nutritional and chronobiological studies in fish.