A prospective study was performed to develop a method for assessing "on-line" error detection and correction during performance of naturalistic action, to determine whether traumatic brain injury (TBI) affects error detection and correction, and to compare actual task performance with verbal self-ratings of performance. Participants included 18 persons who had sustained severe TBI from 34 to 186 days prior to study and who were comparable to controls in their rate of naturalistic action error, along with 18 control subjects chosen to be demographically comparable to subjects with TBI. Subjects performed two different tests of naturalistic action in which they completed everyday activities (eg, wrapping a gift, making toast) at different levels of complexity, as manipulated by the addition of distractor objects, the number of tasks that had to be completed per trial, and other demands on planning and working memory. Using a specially developed coding system, each error on these tasks was scored as to whether the subject corrected it and whether the subject otherwise demonstrated awareness of the error. Error scores were also compared to subjects' responses to a questionnaire in which they rated their own performance on the most challenging level of the naturalistic action test. In general, subjects with TBI corrected and showed awareness of proportionally fewer of their errors when compared to controls. Qualitative patterns for some error types also differed between groups. Despite making more errors than control subjects on the most challenging task, subjects with TBI did not rate themselves as performing more poorly with respect to its cognitive demands. However, for subjects with TBI, the number of errors was correlated with performance ratings on certain questionnaire items. This study showed that error detection and correction can be reliably measured during naturalistic action and appear to be impaired in severe TBI even when the base rate of error is controlled. TBI may affect error detection and correction by reducing, or impairing the allocation of, attentional resources needed for the simultaneous execution and monitoring of routine action.