Existing schemes for victim-offender classification are not standardized, are often multidimensional and contain overlapping categories, and often use vague and incomplete definitions. The proposed coding instrument, called an 'attribute coding form,' breaks down the classification of victim-offender relationships into a series of binary decisions about the presence or absence of an attribute. Attribute categories for victim-offender relationships are neighbors and co-residents, lovers, friends and associates, family, strangers, and other statuses. This instrument not only increases the reliability of classifications but also allows for relationships falling into more than one category. Additionally, this method strengthens the researcher's ability to identify sources of reliability problems. This study compares estimates of the reliability of the Supplementary Homicide Report (SHR) victim-offender classification for Baltimore in 1983 with an estimate of the reliability of the attribute method from a pilot study. Preliminary evidence suggests the attribute approach is more reliable than the SHR. 2 tables, 1 figure, 23 footnotes, and a listing of the attribute coding of family and relatives.