The present paper is based on discharges and suspended particulate matter concentrations from a 9-years high-resolution database for the Garonne River (large plain river) covering contrasted hydrologic years, and a 12-months high frequency sampling for the Nivelle River (small mountainous river). Annual SPM fluxes in the Garonne River range from 0.6 x 10(6) t year(-1) (1997) to 3.9 x 10(6) t year(-1) (1996). In contrast, the Nivelle River transported 11 x 10(3) t year(-1) from December 1995 to December 1996. From the long-term observation of the Garonne River an empirical relation between SPM* (discharge-weighted mean annual SPM concentrations) and annual discharge was established. This relation allows estimating annual SPM fluxes for the Garonne River with less than 30% deviation from reference values for the whole range of mean annual discharge observed during the past decade. Specific (=area-normalized) annual SPM fluxes (YSPM) range from 11 to 74 t km(-2) year(-1) for the Garonne River. Comparison of these results with YSPM of the Nivelle River (69 t km(-2) year(-1) in 1996) suggests that interannual hydrological variations may have a greater impact on fluvial SPM transport than basin-specific parameters. By extracting individual SPM concentrations and corresponding discharge values from the database, different sampling frequencies were simulated and resulting SPM fluxes were then compared to reference fluxes derived from the complete database. If a deviation of simulated flux estimates from reference fluxes lower than +/-20% is accepted, the Garonne River (large plain river) must be sampled at least every 3 days (10 samples per month) and the Nivelle River every 7 h (approx. 100 samples per month). For the Garonne River this minimum sampling frequency is valid for all contrasted hydrologic years of the observation period. Below these minimum sampling frequencies, annual SPM flux estimates may greatly differ from reference fluxes (up to 200%) and there is high probability of systematic underestimation. Consequently, annual SPM flux estimates for the Garonne River derived from the empirical relation (SPM*-annual discharge) are likely to be more satisfactory (errors <30%) than estimates based on sampling frequencies lower than the minimum frequency. These findings underline the need of adapted sampling strategies for erosion assessment, reliable chemical (e.g. nutrients and pollutants) mass balances and characterisation of fluvial transport mechanisms in the world's contrasted watersheds.