This study presents a reliable, short and practical version of the Clock Drawing Test (CDT) for clinical use and examines its diagnostic accuracy in mild Alzheimer's disease versus elderly nonpatients. Clock drawings from 231 participants were scored independently by four clinical neuropsychologists blind to diagnostic classification. The interrater agreement of individual scoring criteria was analyzed and items with poor or moderate reliability were excluded. The classification accuracy of the resulting scoring system - the six-item CDT - was examined. We explored the effect of further reducing the number of scoring items on classification accuracy and estimated classification accuracy associated with performances deviating from the optimal cutoff score. At a cutoff of 5/6, the six-item CDT had a sensitivity (SN) of 0.65 and a specificity of 0.80. Stepwise removal of up to three items reduced SN slightly. Classification accuracy associated with a score of four or less out of six was very high.