Measurements in complex matrices like milk still present a challenge in biosensor development. This is especially important when using a label-free detection method or when measuring low analyte concentrations. The direct optical method reflectometric interference spectroscopy (RIfS) was used for investigating matrix effects in immunoassay development. Furthermore, approaches to reduce these effects have been established. As a model system, the hormone testosterone has been chosen because this immunoassay has been well characterized in buffer. In a first step, the immunoassay for the detection of testosterone in buffer was improved beyond former published results. Therefore, the sensor surface was optimized, resulting in a fivefold lower limit of detection (70.2 ng L(-1)) and limit of quantification (130.0 ng L(-1)). Additionally, the assay time could be reduced to 15 min. Consequently, we used this improved assay to investigate matrix effects of whole pasteurized bovine milk. To minimize these effects, the surface chemistry was adapted and a suitable evaluation method was established, reducing the effects of Tyndall scattering and nonspecific binding to the sensor surface. These improvements allow for very reliable quantitative measurements in milk. The assay developed required no sample pretreatment and allowed for the regeneration of the sensor surface so that calibration could be performed on one chip. The calibration in milk (3.5% fat) resulted in a limit of detection of 94.4 ng L(-1) and a limit of quantification of 229.3 ng L(-1). Furthermore, recovery rates between 70% and 120% could be obtained. Thus, for the first time, an analyte in the matrix milk was successfully quantified with RIfS at low concentrations.