PURPOSE: Existing definitions of clinically important weight loss (WL) in patients with cancer are unclear and heterogeneous and do not consider current trends toward obesity. METHODS: Canadian and European patients with cancer (n = 8,160) formed a population-based data set. Body mass index (BMI) and percent WL (%WL) were recorded, and patients were observed prospectively until death. Data were entered into a multivariable analysis controlling for age, sex, cancer site, stage, and performance status. Relationships for BMI and %WL to overall survival were examined to develop a grading system. RESULTS: Mean overall %WL was -9.7% ± 8.4% and BMI was 24.4 ± 5.1 kg/m(2), and both %WL and BMI independently predicted survival (P < .01). Differences in survival were observed across five categories of BMI (< 20.0, 20.0 to 21.9, 22.0 to 24.9, 25.0 to 27.9, and ≥ 28.0 kg/m(2); P < .001) and five categories of %WL (-2.5% to -5.9%, -6.0% to -10.9%, -11.0% to -14.9%, ≥ -15.0%, and weight stable (± 2.4%); P < .001). A 5 × 5 matrix representing the five %WL categories within each of the five BMI categories was graded based on median survival and prognostic significance. Weight-stable patients with BMI ≥ 25.0 kg/m(2) (grade 0) had the longest survival (20.9 months; 95% CI, 17.9 to 23.9 months), and %WL values associated with lowered categories of BMI were related to shorter survival (P < .001), as follows: grade 1, 14.6 months (95% CI, 12.9 to 16.2 months); grade 2, 10.8 months (95% CI, 9.7 to 11.9 months); grade 3, 7.6 months (95% CI, 7.0 to 8.2 months); and grade 4, 4.3 months (95% CI, 4.1 to 4.6 months). Survival discrimination by grade was observed within specific cancers, stages, ages, and performance status and in an independent validation sample (n = 2,963). CONCLUSION: A robust grading system incorporating the independent prognostic significance of both BMI and %WL was developed.