Estimates of unintentional deaths from carbon monoxide (CO) poisoning can be obtained from national mortality data. We explored ways of accurately estimating CO-related deaths from International Classification of Diseases, 9th Revision (ICD-9) coded U.S. mortality data. We evaluated and identified CO-related ICD-9 codes and created five classes of codes for case ascertainment that represented a continuum of the degree of certainty that the ICD-coded death was truly CO-related. We conducted single (underlying) cause-of-death and multiple-cause-of-death analysis using 20 years of data (1979-1998), and calculated sensitivity and positive predictive value using different criteria for case ascertainment. Single-cause analysis provided accurate estimates only when we used CO-exclusive E-codes, however this method failed to identify approximately one third of the CO-related deaths over the study period. Single-cause analysis overestimated the number of CO-related deaths when we used E-codes that were not exclusive to CO exposure. Identification of true CO-related deaths required multiple cause-of-death analysis and use of the CO nature-of-injury code, N986, to confirm suspected cases. Sensitivity of N986 was 99.5%, and positive predictive value of the individual E-codes in single cause-of-death analysis ranged from 1.5% to 92%. Estimating CO-related deaths from ICD-coded data requires a thorough understanding of the ICD codes, coding rules, and of the limitations imposed by case selection criteria and single cause-of-death analysis.