Manning's roughness coefficient is estimated for a gravel-bed river reach by field measurement on water level and discharge and applicability of various methods used for estimation of roughness coefficient is evaluated. Results show that roughness coefficient tends to decrease with increasing discharge and relative depth, and over a certain range it appears to become constant. Comparison of roughness coefficients by field measurement with ones estimated by several previous methods shows that though they give approximate roughness coefficient values for relatively large discharge, there seem to be rather high uncertainty due to difference of resultant values. For this, uncertainties related to roughness coefficient are analyzed in terms of change in computed variables. In average, a +20% uncertainty in roughness coefficient causes a 7% increase in water level and an 8% decrease in velocity, but they may be about a 15% increase and as much decrease for a certain cross-section in the study reach. Finally, validity of roughness coefficient estimation based on field measurement is examined. A 10% error in discharge measurement may lead to over 10% uncertainty in roughness coefficient estimation, but corresponding uncertainty in computed water depth and velocity is reduced to approximately 5%. Conversely, necessity for roughness coefficient estimation by field measurement may be confirmed.