Halo effects in rating specific pieces of work, as in educational grading, have received little attention. Grades awarded by 2 independent graders to undergraduate projects were analyzed with a correlated uniqueness model. Grades showed substantial halo despite being awarded by expert assessors at the time of reading the work. There was greater halo between different grades applying to the same section of the project than between grades applying to different sections. Supervisors who had regular contact with the student whose work they were grading showed no more halo than other graders. More reliable graders showed less within-section halo than graders of lower reliability but equal between-sections halo. The halo effects observed cannot be entirely attributable to a unitary general impression.