- We describe an algorithm to decompose deep images of active galactic nuclei into host galaxy and nuclear components. Currently supported are three galaxy models: a de Vaucouleurs spheroidal; an exponential disc; and a two- component disc + bulge model. Key features of the method are: (semi-)analytic representation of a possibly spatially variable point spread function; full two-dimensional convolution of the model galaxy using gradient-controlled adaptive subpixelling; and a multiple iteration scheme. The code is computationally efficient and versatile for a wide range of applications. The quantitative performance is measured by analysing simulated imaging data. We also present examples of the application of the method to small test samples of nearby Seyfert 1 galaxies and quasars at redshifts z < 0.35
