We assess how the size of the Helley-Smith (HS) bed load sampler nozzle affects the accuracy of bed load sampling. Semitheoretical considerations show that the larger grains resident on the streambed can influence the sample either by blocking the sampler entrance or by causing the sampler to rest in a ``perched'' position. Probabilities for interference can be derived from the distribution of grain sizes but they do not capture the actual complexity of the influence of the bed on sampler performance. We therefore make an empirical comparison of sediment trapped by HS samplers with 76- and 152-mm intakes during floods in the gravel bed lower Ebro River. Most bed load rates appeared higher when sampled with the HS152. The largest clasts collected by the HS76 also tend to be smaller than those obtained with the HS152 at the same flow. Analyzing paired bed load samples, we find the probability of a bed load sample collected with the HS152 to be biased is around 43% in the conditions of the present study, whereas 65% of samples were biased when obtained with the HS76. The analysis emphasizes the influence of bed material texture over sampler performance and demonstrates that the use of samplers with intake size much larger than bed grain size (i.e., ~5D) will increase the accuracy of bed load grain size distributions and the precision of annual load estimates in gravel bed rivers.