In a small mountain stream (Erlenbach, Alptal, central Switzerland), water samples were taken during 15 years by twodifferent methods: weekly-bulked, discharge-proportional sampling and fortnightly grab (instantaneous) sampling. Comparedwith continuous, in-stream measurements of the electrical conductivity, the discharge-proportional sampling proved to givevery accurate and precise estimates of solute fluxes. Relatively to discharge-proportional samples, integration methods based onthe grab samples resulted in more or less biased fluxes of: Ca, Mg, K, Na, SO₄^2-, Cl^-,NO₃^- and electrical conductivity. Unweighted means were all positively biased due to concentrations decreasing with the discharge. Means weighted bydischarges averaged fortnightly were less positively biased. On the other hand, weighing by the instantaneous discharge gavenegative biases. Fluxes calculated from regressions of concentrations against discharge were also negatively biased, but theaccuracy could partly be improved by including further factors into the regressions, especially the seasonality. If both grab anddischarge-proportional samples are taken in parallel for some time, it is possible to calibrate the fluxes calculated from the firstagainst the second. Calibration periods of 1–7 years were tested and found to improve the accuracy in 8 other years used forvalidation. Five years were sufficient to achieve unbiased regression estimates and a precision better than 10%. Replacing a lessaccurate grab sampling scheme by a discharge-proportional sampling should thus be done with a transition period during whichboth methods are run in parallel, 5 years being advisable for most solutes and for small streams.