During a 16 day period with pronounced snowmelt via surface runoff, high water concentrations (usually 0·4–0·5 mg l−1) of dissolved molybdate-reactive phosphorus (MRP) were detected in surface runoff water from a clay soil of illite type. Other phosphorus fractions defined were: phosphorus in particles with a higher settling coefficient than 80 000 S (SPP); colloidal phosphorus caught on filters with a pore size of 0·2 µm but with a smaller settling coefficient (CPP); and dissolved phosphorus not reacting with molybdate (DUP). The order of concentrations was MRP > SPP > CPP > DUP. Nearly identical amounts of MRP, CPP, and DUP (in total 0·3 kg ha−1) were lost from a grass–clover ley and a ploughed soil. However, more of the heavier phosphorus-containing material was lost from the ploughed area. In drainpipe water, CPP was the largest fraction (28%), and in stream water from mixed arable/forest land, MRP dominated (33%). Loss on ignition of the settling material slowly decreased from 10 to 8% (dry weight) during the snowmelt period. Total phosphorus concentrations in the material followed the runoff pattern, with slightly higher phosphorus concentrations during fast runoff. The large amounts of readily dissolved or colloidal-bound phosphorus (70–80%) transported from this clay soil during snowmelt are discussed with regard to the use of grass buffer strips as a measure against phosphorus losses from arable land. Copyright © 2002 John Wiley & Sons, Ltd.