Two methods were studied for selectively measuring the on-chip absorbance of trace sulfate analytes in droplets on a digital microfluidics (DMF) platform. In one method, the direction of measurement was perpendicular to the flat upper and lower surfaces of the DMF platform (vertical), and in the second method, the measurement direction was parallel to the DMF platform surfaces (horizontal). The channel height or the vertical light path length was 0.24 mm, and the droplet diameter was 1 mm. The DMF system employed a silicone oil transport medium whereby a thin, non-uniform oil layer formed between the droplet and the upper/lower plates which was unstable, resulting in randomly formed local oil lenses. The mobile oil lenses caused vertical absorbance measurement errors and uncertainties. The effects of the oil lenses were verified by simulation. Horizontal absorbance measurements were taken with embedded optical fibers (0.2 mm in diameter) aligned over the bottom chip surface in contact with the sides of the droplet, resulting in a horizontal light path length approximately three times that of the vertical light path. Because no oil lenses could form on the droplet’s sides, the stability of repeated horizontal measurements outperformed repeated vertical measurements made on the same droplet and on multiple droplets actuated into the measurement positions. Comparisons were based on measurement standard deviations and limits of detection (LOD). The following LODs and measurement standard deviations were achieved for horizontal measurements of multiple sulfate concentrations in 1.5 µl droplets: 7 ppm for sulfate (0.3–2.7%) and an R2 value of 0.957 from a least square data fit. Measurements on a commercial plate reader gave comparable results (200 µl liquid in each well, LOD equals 11 ppm, CV equals to 0.2–4%), even though the absorbance path was larger (0.7 mm). This LOD value means that the chip could detect 10.5 ng of sulfate. LOD values on vertical measurements were also similar, but large measurement errors from numerous outlier points yielded an R2 value of 0.735 and large average measurement standard deviations (36%).