Context. Asteroid (21) Lutetia was the target of the ESA Rosetta mission flyby in 2010 July. Aims. We seek the best size estimates of the asteroid, the direction of its spin axis, and its bulk density, assuming its shape is well described by a smooth featureless triaxial ellipsoid. We also aim to evaluate the deviations from this assumption. Methods. We derive these quantities from the outlines of the asteroid in 307 images of its resolved apparent disk obtained with adaptive optics (AO) at Keck II and VLT, and combine these with recent mass determinations to estimate a bulk density. Results. Our best triaxial ellipsoid diameters for Lutetia, based on our AO images alone, are a × b × c = 132 × 101 × 93 km, with uncertainties of 4 × 3 × 13 km including estimated systematics, with a rotational pole within 5The AO model fit itself has internal precisions of 1 × 1 × 8 km, but it is evident both from this model derived from limited viewing aspects and the radius vector model given in a companion paper, that Lutetia significantly departs from an idealized ellipsoid. In particular, the long axis may be overestimated from the AO images alone by about 10 km. Therefore, we combine the best aspects of the radius vector and ellipsoid model into a hybrid ellipsoid model, as our final result, of diameters 124 ± 5 × 101 ± 4 × 93 ± 13 km that can be used to estimate volumes, sizes, and projected areas. The adopted pole position is withinConclusions. Using two separately determined masses and the volume of our hybrid model, we estimate a density of 3.5 ± 1.1 or 4.3 ± 0.8 g cm −3 . From the density evidence alone, we argue that this favors an enstatite-chondrite composition, although other compositions are formally allowed at the extremes (low-porosity CV/CO carbonaceous chondrite or high-porosity metallic). We discuss this in the context of other evidence.