In 46 epileptic patients the range of the unbound fraction of phenytoin in plasma measured by ultrafiltration (at 37 degrees C) and tracer-labelling with [14C]-phenytoin was 6.7%-33.3% with a median of 11.9%. The total and unbound phenytoin plasma concentrations were significantly correlated (r = 0.93, P less than 0.001), but in six patients the unbound concentration fell on or outside the 90% predictability limits for a single value. In all patients the unbound concentration reflected the clinical status of the patient equally or better than the total concentration. An inverse relationship was found between the plasma albumin concentration (within the normal reference range) and the phenytoin unbound fraction (r = -0.83, P less than 0.001) indicating that plasma albumin concentration is one of the important overall determinants of phenytoin protein binding. Saliva and plasma unbound phenytoin concentrations were significantly correlated (r = 0.98, P less than 0.001) but both collection of plasma samples and preparation of plasma ultrafiltrate using the Amicon micropartition system are simpler than collection and processing saliva, and interpretation of plasma unbound concentration does not require allowance for potential contamination. The additional value of the unbound phenytoin concentration in a clinically significant number of individuals would justify routine measurement of unbound phenytoin concentration in monitoring therapy, once available simplified methodology has been adequately characterised.