Raw milk (2,710 kg) was separated at 4 degrees C, the skim milk was pasteurized (72 degrees C, 16 s), split into 3 batches, and microfiltered using pilot-scale ceramic uniform transmembrane pressure (UTP; Membralox model EP1940GL0.1microA, 0.1 microm alumina, Pall Corp., East Hills, NY), ceramic graded permeability (GP; Membralox model EP1940GL0.1microAGP1020, 0.1 microm alumina, Pall Corp.), and polymeric spiral-wound (SW; model FG7838-OS0x-S, 0.3 microm polyvinylidene fluoride, Parker-Hannifin, Process Advanced Filtration Division, Tell City, IN) membranes. There were differences in flux among ceramic UTP, ceramic GP, and polymeric SW microfiltration membranes (54.08, 71.79, and 16.21 kg/m2 per hour, respectively) when processing skim milk at 50 degrees C in a continuous bleed-and-feed 3x process. These differences in flux among the membranes would influence the amount of membrane surface area required to process a given volume of milk in a given time. Further work is needed to determine if these differences in flux are maintained over longer processing times. The true protein contents of the microfiltration permeates from UTP and GP membranes were higher than from SW membranes (0.57, 0.56, and 0.38%, respectively). Sodium-dodecyl-sulfate-PAGE gels for permeates revealed a higher casein proportion in GP and SW permeate than in UTP permeate, with the highest passage of casein through the GP membrane under the operational conditions used in this study. The slight cloudiness of the permeates produced using the GP and SW systems may have been due to the presence of a small amount of casein, which may present an obstacle in their use in applications when clarity is an important functional characteristic. More beta-lactoglobulin passed through the ceramic membranes than through the polymeric membrane. The efficiency of removal of serum proteins in a continuous bleed-and-feed 3x process at 50 degrees C was 64.40% for UTP, 61.04% for GP, and 38.62% for SW microfiltration membranes. The SW polymeric membranes had a much higher rejection of serum proteins than did the ceramic membranes, consistent with the sodium-dodecyl-sulfate PAGE data. Multiple stages and diafiltration would be required to produce a 60 to 65% serum protein reduced micellar casein concentrate with SW membranes, whereas only one stage would be needed for the ceramic membranes used in this study.