We present a new approach for spatial interpolation of experimental raindrop size distribution (DSD) spectra. The DSD is fundamental to the study and understanding of precipitation and its monitoring and modelling. It is measured insitu using disdrometers at point locations. Disdrometers provide a (non-parametric) DSD spectrum in which drop concentrations are provided per class of drop diameter. Our approach uses geostatistics to estimate the same non-parametric DSD at unmeasured locations. Non-stationarity due to intermittency is taken into account through estimation of the dry drift of drop concentrations, using a rain occurrence field. Principal component analysis is used to express the DSD spectra in terms of uncorrelated components that can be interpolated independently at a requested point. These interpolated components can then be recombined into the full DSD. Estimation uncertainty for the interpolated DSD spectra is provided. Because all bulk rainfall variables can be calculated from the DSD and the entire DSD is estimated, the technique effectively interpolates all bulk variables at once. Leave-one-out testing shows that the technique estimates the DSD with minimal bias, and it is shown that the technique can be easily adapted to perform stochastic simulation of the non-parametric DSD.