This study aimed to explore the effect of curve features (radius and type) along rural, two-lane, undivided (RTU) roads on the risk of single-motorcycle crash occurrence and to develop associated crash modification factors (CMFs). A matched case-control study was used to address three typical issues (low sample-mean, aggregation bias over years, and uncontrolled confounding factors) in traditional cross-sectional analyses of motorcycle crash data by matching controls (entities without crash records) to cases (entities with crash records) in year, Average Annual Daily Traffic (AADT), and segment length. A total of 1,601 cases and 16,010 controls (for 2005–2015) were collected from Florida RTU roads. Ten controls were matched to each case through random sampling, which made the power of design archive to 98%. A conditional logistic regression model was developed based on the case-control pairs to calculate the odds ratios of explanatory variables. Based on model estimation results, the relative risks of sharp curves (radius < 1,500 ft) with non-reverse design, sharp curves with reverse design, moderate curves (1,500 – 3,000 ft) with reverse design, moderate curves without reverse design, slight curves (3,000 – 8,000 ft), and flat curves (8,000 – 20,000 ft) were estimated as 4.92 times, 3.21 times, 2.62 times, 2.00 times, 1.88 times, 1.62 times, respectively, as high as that on straight segments. All CMFs were significant at a 95% confidence level. In addition, motorcycle rider safety compensation behaviors were found to be associated with several factors, including reverse design on sharp curves, narrow surface and shoulder widths, paved shoulder, and poor pavement condition.