The volume of synovial fluid in a joint correlates inversely with plasma colloid osmotic pressure (COP). The inferred influence of plasma osmotic forces on trans-synovial flow was investigated directly here, in isolated perfused hindquarters of sixteen rabbits. Flow of intra-articularly infused Krebs solution across the synovial lining of the cannulated knee was recorded at controlled intra-articular pressure (18 cmH2O). Colloid osmotic pressure in the synovial microcirculation was varied by perfusion with oxygenated red cells resuspended in albumin solution or plasma from an extra-corporeal system at constant perfusion pressure. Studies in vitro showed that the COP versus concentration curve for commercial bovine albumin samples was variable and not reliably described by a widely used polynomial. The rate of trans-synovial absorption Qs was a positive linear function of intravascular COP pi p (r = 0.936, P less than 0.001, n = 83). The average slope dQs/d pi p was 0.20 microliter min-1 cmH2O-1 (S.E. +/- 0.01 microliter min-1 cmH2O-1), the slope depending on hydraulic conductance and osmotic reflection coefficient. Trans-synovial flow was a negative linear function of synovial capillary pressure (Pc). Absolute slope dQs/d pi p was on average only 78% of dQs/dPc in the same joint. The osmotic reflection coefficient of the blood-joint barrier to serum albumin was estimated from these slopes as 0.78-0.81 (S.E.M. +/- 0.06). Vascular perfusion with a hyperosmolar solution of glucose, sucrose or NaCl generated a transient, rapidly decaying osmotic absorption from the joint cavity, with a half-life of 17-60 s. A reversed osmotic transient occurred on reperfusion with isotonic fluid. It was concluded that the blood-joint barrier, which comprises fenestrated endothelium and synovial intima, approximates to an imperfect semipermeable membrane for albumin solutions, justifying the application of Starling's hypothesis to trans-synovial flow. For small solutes the tissues form a highly permeable but nevertheless slightly osmotically reflective membrane.