This paper, presented at a meeting on Systematic Reviews in 1993, emphasizes the importance of investigating sources of heterogeneity in meta-analyses to enhance their scientific and clinical relevance. The author distinguishes between clinical and statistical heterogeneity, illustrating the need for careful exploration of these differences through examples from meta-analyses of epidemiological studies and clinical trials. Clinical heterogeneity refers to differences in patient characteristics, interventions, and outcomes across studies, while statistical heterogeneity is the variation in results that cannot be attributed to chance alone. The paper argues that a sensible investigation of heterogeneity should focus on specific contrasts between trials rather than relying solely on overall statistical tests, as this approach is more powerful and clinically relevant. Examples from meta-analyses of serum cholesterol concentration and its reduction are provided to demonstrate how exploring heterogeneity can lead to more nuanced and accurate conclusions. The discussion highlights the importance of considering factors such as age, socioeconomic status, and treatment duration in meta-analyses to avoid overinterpretation and ensure the robustness of the results.This paper, presented at a meeting on Systematic Reviews in 1993, emphasizes the importance of investigating sources of heterogeneity in meta-analyses to enhance their scientific and clinical relevance. The author distinguishes between clinical and statistical heterogeneity, illustrating the need for careful exploration of these differences through examples from meta-analyses of epidemiological studies and clinical trials. Clinical heterogeneity refers to differences in patient characteristics, interventions, and outcomes across studies, while statistical heterogeneity is the variation in results that cannot be attributed to chance alone. The paper argues that a sensible investigation of heterogeneity should focus on specific contrasts between trials rather than relying solely on overall statistical tests, as this approach is more powerful and clinically relevant. Examples from meta-analyses of serum cholesterol concentration and its reduction are provided to demonstrate how exploring heterogeneity can lead to more nuanced and accurate conclusions. The discussion highlights the importance of considering factors such as age, socioeconomic status, and treatment duration in meta-analyses to avoid overinterpretation and ensure the robustness of the results.