The scale of fluctuation (SOF) is an important parameter for the characterization of soil spatial variability, and different fluctuation scales can be used to simulate different degrees of the spatial variability of soil parameters. Based on the random finite element method (RFEM) and Monte-Carlo simulation, the heterogeneous soil mass is modeled assuming that the elastic modulus follows a lognormal distribution, by which the settlement deformation of a circular foundation under different fluctuation scales is studied. The results show that the central settlement is well fitted by a log-normal distribution, and the differential settlement is closely approximated by a normal distribution. With the increase of SOF, the mean value of central settlement and differential settlement are basically constant. The standard deviation of central settlement first increases and then tends to level off. In contrast, the standard deviation of differential settlement first increases and then decreases; it reaches the maximum when the SOF is close to the diamenter of the circular foundation, at which the average absolute differential settlement also reaches the peak. This SOF is the most unfavorable scale and is termed as the worst-case SOF. The probability of foundation settlement can be conservatively estimated by assuming such a worst-case SOF when the site survey data are relatively insufficient to define a reliable SOF.