Dry bulk density expresses the ratio of the dry mass of a soil sample to its total volume (solid and pores). Fractal models of soil structure predict a power-law relationship between mass and total volume or, equivalently, between bulk density and sample volume. Measurement of total volume of irregular porous objects such as aggregates or clods is generally difficult and time consuming because it relies on defining the volume of a porous material. The objective of this work was to study the effect of sample size on the estimate of mass fractal dimension (from bulk density-diameter in a log-log representation) related to those obtained with 100% of the experimental data. Differential buoyancy in two non-mixing liquids of contrasting density was used to measure aggregate volume, by saturating aggregates with the liquid of lower density followed by submersion in the liquid of higher density. Aggregates sampled from the A horizon of three soils (Ryders n=72, Gladstone n=83 and Holmdel n=72), each under two contrasting management situations (wooded and cultivated) were used in the calculations. Twenty subsets of experimental data, containing 30, 40 and 50% of the original data, were fitted by linear regression to estimate 20 mass fractal dimension values. The subsets were constituted by randomly selected samples from the complete data set for each soil and management situation. In total we obtained 360 fractal dimension values. The mass fractal dimension of the entire data set and of subsets encompassing 30, 40 and 50% of the data sets were fitted to a power-law relation in a log-log transformation. A multifactor analysis of variance was used to compare fractal dimensions from the different data set. The results were not uniform across soils and management types. Ryders soil showed that the management situation affect the value of the fractal dimension. On the contrary, there's not effect of the size of the subset of sample on the fractal dimension. For Gladstone there's neither effect of management situation nor size of the subset of sample on the fractal dimension. For Holmdel the analysis of variance showed that both management situation and size of the subset of sample gave significant statistical differences with reference to fractal dimension values. The various estimates of fractal dimensions will be discussed by considering their impact on the prediction of hydraulic properties and ultimately on the modeling of water flow through soil.