Background Variation in floral shapes has long fascinated biologists and its modelling enables testing of evolutionary hypotheses. Recent comparative studies that explore floral shape have largely ignored 3D floral shape. We propose quantifying floral shape by using geometric morphometrics on a virtual3D model reconstructed from 2D photographical data and demonstrate its performance in capturing shape variation. Methods This approach offers unique benefits to complement established imaging techniques (i) by enabling adequate coverage of the potential morphospace of large and diverse flowering-plant clades; (ii) by circumventing asynchronicity in anthesis of different floral parts; and (iii) by incorporating variation in copy number of floral organs within structures. We demonstrate our approach by analysing 90 florally-diverse species of the Southern African genus Pelargonium (Geraniaceae). We quantify Pelargonium floral shapes using 117 landmarks and show similarities in reconstructed morphospaces for nectar tube, corolla (2D datasets), and a combined virtual3D dataset. Results Our results indicate that Pelargonium species differ in floral shape, which can also vary extensively within a species. PCA results of the reconstructed virtual3D floral models are highly congruent with the separate 2D morphospaces, indicating it is an accurate, virtual, representation of floral shape. Through our approach, we find that adding the third dimension to the data is crucial to accurately interpret the manner of, as well as levels of, shape variation in flowers.