The development of fiber-optic biosensors requires that a biorecognition element and a fluorescent reporter group be immobilized at or near the surface of an optical element such as a planar waveguide or optical fiber. In this study, we examined a model biorecognition element–reporter group couple consisting of human serum albumin that was site-selectively labelled at Cys 34 with iodoacetoxy-nitrobenzoxadiazole (HSA–NBD). The labelled protein was encapsulated into sol–gel derived materials that were prepared either as monoliths, as beads that were formed at the distal tip of a fused silica optical fiber, or as thin films that were dipcast along the length of a glass slide or optical fiber. For fiber-based studies, the entrapped protein was excited using a helium–cadmium laser that was launched into a single optical fiber, and emission was separated from the incident radiation using a perforated mirror beam-splitter, and detected using a monochromator–photomultiplier tube assembly. Changes in fluorescence intensity were generated by denaturant-induced conformational changes in the protein or by iodide quenching. The analytical parameters of merit for the different encapsulation formats, including minimum protein loading level, response time and limit-of-detection, were examined, as were factors such as protein accessibility, leaching and photobleaching. Overall, the results indicated that both beads and films were suitable for biosensor development. In both formats, a substantial fraction of the entrapped protein remained accessible, and the entrapped protein retained a large degree of conformational flexibility. Thin films showed the most rapid response times, and provided good detection limits for a model analyte. However, the entrapment of proteins into beads at the distal tip of fibers provided better signal-to-noise and signal-to-background ratios, and required less protein for preparation. Hence, beads appear to be the most viable method for interfacing of proteins to optical fibers.