In this paper automatic sensor identification of sensor classes within a high-density randomized array, without a priori knowledge of sensor locations, is demonstrated. Two different fluorescence-based sensor types, with hundreds of replicates each, were randomly distributed into an optical imaging fiber array platform. The sensor element types were vapor-sensitive microspheres with the environmentally-sensitive fluorescent dye Nile Red adsorbed on their surface. Nile Red undergoes spectral changes when exposed to different microenvironmental polarity conditions, e.g. microsphere surface polarity or odor exposure. These reproducible sensor spectral changes, or sensor-response profiles, enable sensors within a randomized array to be grouped into categories by optical decoding methods. Two computational decoding methods (supervised and unsupervised) are introduced; equal classification rates were achieved for both. By comparing sensor responses from a randomized array with those obtained from known (control) arrays, 587 sensors were correctly classified with 99.32% accuracy. Although both methods were equally effective, the unsupervised method, which uses sensor response changes to odor exposure, is a better decoding model for the vapor-sensitive arrays studied, because it relies only on the odor-response profiles. Another decoding technique employed the emission spectra of the sensors and is more applicable to other types of multiplexed fluorescence-based arrays and assays. The sensor-decoding techniques are compared to demonstrate that sensors within high-density optical chemosensor arrays can be positionally-registered, or decoded, with no additional overhead in time or expense other than collecting the sensor-response profiles.