Software documentation for the novice user typically must try to achieve at least three goals: to support basic knowledge and skills development; to prevent or support the handling of mistakes, and to support the joint handling of manual, input device and screen. This paper concentrates on the latter goal. Novice users often experience split-attention problem due to the need to (almost) simultaneously attend to different media. Existing research indicates that split-attention problems can be prevented or reduced by the presence of screen images in the manual. Research is yet unclear about the optimal design of these pictures. This study examines three design styles. Forty-eight novice users received one of the three manual based on these styles. The manuals were an introduction to Windows 95. The users of the most successful manual needed 25% less training time and had a 60% better retention. The most important characteristics of the design style of this manual were its use of full screen images (instead of partial ones) and a two-column lay-out in which the instructions and screen images were presented side-by-side in a left-to-right reading order. The discussion focuses on the tension that exists between theory and practice. Special attention is given to the contributions of a taxonomy of screen images and cognitive load theory.