Questionnaire and biochemical measures of smoking were studied in 211 hospital outpatients. Eleven different tests of smoke intake were compared for their ability to categorize smokers and nonsmokers correctly. The concentration of cotinine, whether measured in plasma, saliva, or urine, was the best indicator of smoking, with sensitivity of 96-97 per cent and specificity of 99-100 per cent. Thiocyanate provided the poorest discrimination. Carbon monoxide measured as blood carboxyhaemoglobin or in expired air gave sensitivity and specificity of about 90 per cent. Sensitivities of the tests were little affected by the presence among the claimed nonsmokers of a group of 21 "deceivers" who concealed their smoking. It is concluded that cotinine is the measure of choice, but for most clinical applications carbon monoxide provides an acceptable degree of discrimination and is considerably cheaper and simpler to apply.