Adapting state-space models (SSMs) to telemetry data has been helpful for dealing with location error and for modeling animal movements. We used a combination of two hierarchical Bayesian SSMs to estimate movement pathways from Argos satellite-tag data for 15 juvenile loggerhead turtles (Caretta caretta) in the western Mediterranean Sea, and to probabilistically assign locations to one of two behavioral movement types and relate those behaviors to environmental features. A Monte Carlo procedure helped propagate location uncertainty from the first SSM into the estimation of behavioral states and environment--behavior relationships in the second SSM. Turtles using oceanic habitats of the Balearic Sea (n = 9 turtles) within the western Mediterranean were more likely to exhibit "intensive search" behavior as might occur during foraging, but only larger turtles responded to variations in sea-surface height. This suggests that they were better able than smaller turtles to cue on environmental features that concentrate prey resources or were more dependent on high-quality feeding areas. These findings stress the importance of individual heterogeneity in the analysis of movement behavior and, taken in concert with descriptive studies of Pacific loggerheads, suggest that directed movements toward patchy ephemeral resources may be a general property of larger juvenile loggerheads in different populations. We discovered size-based variation in loggerhead distribution and documented use of the western Mediterranean Sea by turtles larger than previously thought to occur there. With one exception, only individuals > 57 cm curved carapace length used the most westerly basin in the Mediterranean (western Alborán Sea). These observations shed new light on loggerhead migration phenology.