A UNIX shell (csh) was modified to automatically correct misspelled command lines. The design and implementation were not easy, contrary to the opinions of previous researchers. After implementation, the shell was given to 21 users for evaluation. Their comments and performance suggest that any change to a familiar system may overshadow potential benefits of a new system. Designers of spelling correctors must be extremely cautious in order to produce a satisfactory system. Minimal distraction from the user's task should be a prime goal in designers of correction facilities. Spelling corrector filters and methods of testing filters are discussed.