This paper describes an experimental study of the effect of atrium roof form, internal structural obstructions and sky type (overcast and partly cloudy) upon the daylight levels in atrium buildings. Simultaneous daylight measurements were made in two identical atrium models under real skies. One atrium model had an A-frame glazed roof with obstructions created by white structural elements. The second model had a flat glass roof with no obstructions. For overcast skies it was found that applying the plan area obstruction correction factor to daylight levels usually underestimated the loss of light due to the obstructions, typically by 10-15%. For the partly cloudy sky the effects of roof shape and obstruction on atrium daylight levels were complex. The inclined glazed surfaces and structural elements could reflect the bright skylight either away from or into the atrium. Under some circumstances the presence of the obstructions in the roof increased the illuminance levels on north-and east-facing atrium surfaces (compared with the clear flat roof), while for south- and west-facing surfaces illuminance levels decreased.