Methods for on-farm extraction of low-concentration (minor) proteins from raw whole bovine milk directly after milking were explored. These minor proteins have high commercial value. Lactoferrin (LF) and lactoperoxidase (LP) were used as model proteins for extraction using cation exchange chromatography. Laboratory fractionations showed that milk could be processed by conventional column chromatography without excessive column backpressures if resin with large particles sizes were used and the temperature was high enough so fat in the milk was malleable; ideally the milk should be near the secretion temperature of 37oC. Processing parameters such as equilibrium and dynamic capacities were determined for SP Sepharose ™ (GE Healthcare Technologies) and Bio Rex 70 (BioRad Laboratories) resins. SP Sepharose Big Beads (SP BB) were found to be more suitable than BR 70, for raw whole milk processing due to the larger size (200 um). Design considerations showed that column chromatography was not the most practical method for on-farm processing of fresh, raw whole milk. Trials with a single-stage stirred tank showed that SP BB resin could extract up to 65% of LF (initial LF concentration of 0.5 mg/mL) with a 10-minute adsorption time. The composite non-linear (CNL) model of Rowe et al. (1999) was used to describe LF uptake by SP BB resin in raw whole milk with initial LF concentrations of 0 to 1.0 mg/mL and resin:milk volume ratios of 0.010, 0.012, 0.017 and 0.024 over 45-minute contact times. The CNL model could be used to predict LF yields if initial feed concentration, milk and resin volumes, and contact times were known. Laboratory extractions showed that processing did not significantly affect bulk milk composition (fat, protein, lactose and total solids), indicating that the milk could be used for conventional processing after the minor proteins had been extracted. Resin cleaning and regeneration studies, using a procedure similar to that recommended by the resin supplier, showed that the Sepharose resin had not degraded and there was no significant decrease in binding capacity after 50 extraction cycles. A Protein Fractionation Robot (PFR) prototype based on a single-stage stirred tank and the operating parameters obtained from the laboratory trials was designed, assembled and coupled to an Automated Milking System (AMS) to process fresh, raw whole milk from individual cows immediately after milking. The LF and LP extracted from the milk from 16 individual cows were 19.7 - 55.2% (35.6 10.2%) and 21.2 - 99.5% (87.1 12.0%) respectively. Generally, higher extraction levels were obtained at higher resin:milk ratios. The amount of LF extracted on-farm agreed within 14.1 9.8% of those predicted by the CNL model, with predicted values generally being higher. The experimental on-farm adsorption values were calculated using data of LF recovered after elution, so differences between actual and predicted values may be due to losses during post-adsorption processing. Economic feasibility studies, based on experimental data from the PFR and realistic wholesale prices for LF and LP ($400 and $150/kg respectively) showed that PFR-based processing is economically viable if the farmer is paid for the LF and LP produced as well as the bulk milk. This system would have a payback period of approximately five years and an internal rate of return of 14.5%. Further case studies determined the sensitivity of the economics to various operating parameters and value/cost assumptions, including producing recombinant human protein from transgenic bovine milk. These studies showed that the higher the value of the processed raw milk, the higher the absorptive capacity of the resin, and the higher the value of the extracted protein, the more favourable the economics. In the extreme case of producing a very high value therapeutic protein (e.g. $20 000), the payback period could be as low as 0.3 years, with an internal rate of return of 818%.