Precise non-contact micrometer-scale metrology under reflected-light imaging faces contrast loss and edge blurring, which degrade the performance of classical edge detectors. Fresnel diffraction has been widely used in transmission-mode metrology. However, its application in reflective-mode edge localization remains underexplored. In this work, we introduce a reflective-mode Fresnel diffraction approach that leverages dynamic multi-threshold scanning and profile averaging to achieve both high accuracy and robustness to illumination variation. First, a normalized intensity threshold (bI₀, where I₀ is the peak image intensity) is swept from 0.35 to 0.95 in steps of 0.05 to identify precise edge coordinates (x₁, x₂) from diffraction fringes. Second, the diameter D = x2 – x1 is averaged over N = 100 horizontal profiles to reduce the standard error by approximately \(\sim \,1/\sqrt N\) . . Experiments on a 0.29 mm reflective metal cylinder imaged with a 10 MP CCD microscope and ring illumination at 980, 1890, and 3250 lx demonstrate absolute diameter errors below 0.005 mm at optimal thresholds (b = 0.6 for 980/1890 lx, b = 0.7 for 3250 lx) and errors under 0.01 mm across a robust threshold window (b = 0.6–0.7). Compared with single-threshold methods, our approach improves measurement stability without requiring interferometric hardware or complex calibration. These results establish a compact, cost-effective route to reflective Fresnel diffraction metrology achieving sub-2% accuracy. They further suggest future work on adaptive threshold selection, multi-material validation, and comprehensive uncertainty budgeting.