Objective: Answering “I don’t know” (DK) to survey questions that assess risk perceptions is common and occurs more often among disenfranchised groups. Because these groups also are at greater disease risk, statistically omitting or recoding DK responses may disproportionately exclude or misrepresent responses from marginalized groups and misinform intervention efforts. Because little is known about how the DK response is related to health behaviors, we examined whether the relation between DK response and colorectal cancer (CRC) screening behavior differed, depending on the construct queried (knowledge vs. beliefs). Method: Data from the Awareness and Beliefs about Cancer survey of U.S. adults aged 50 years or older (N = 1406) were used. Cancer symptom knowledge, cancer beliefs, and CRC screening history were assessed. DK response options were not offered but were recorded if spontaneously provided. Logistic regression tested whether DK response on knowledge and belief scales were associated with CRC screening. Results: Adjusting for demographic factors, responding DK on ≥1 belief item was associated with lower odds of having been screened for CRC in the last 5 years, p < .001, whereas DK response on ≥1 knowledge item was unassociated with screening, p = .48. DK responders to belief items were even less likely to have been screened than participants with negative beliefs about cancer, p = .011. Conclusion: DK responses are differentially associated with CRC screening, depending on the construct queried, suggesting different mechanisms may underlie DK responding or its relationship with screening. Addressing construct-specific causes of DK response may improve survey research and the interventions based on them.