Constraints on the possible shape and clustering, as well as optical properties, of grains responsible for the 2175Å interstellar extinction feature (interstellar UV bump) are discussed. These constraints are based on the observation that the peak position of the interstellar UV feature is very stable (variations <~1%), that the large variations in width (<~25%) are uncorrelated with the peak position except for the widest bumps, and that the shape of the feature is described extremely well by a Drude profile. The UV extinction of small graphite grains is computed for various clustering models involving Rayleigh spheres. It is shown that compact clusters qualitatively satisfy the above observational constraints, except that the peak position falls at the wrong wavelength. As an alternative to graphite to model the optical properties of the interstellar UV feature carrier, a single-Lorentz oscillator model is considered, in conjunction with a clustering model based on clusters of spheres. Intrinsic changes in the peak position and width are attributed to variations in chemical composition of the grains, impacting upon the parameters of the Lorentz oscillator. Further broadening is attributed to clustering. These models are shown to satisfy the above observational constraints. Furthermore, the correlated shift of peak position with increased width, observed for the widest interstellar UV features, is reproduced. Models involving a second Lorentz oscillator to reproduce the FUV rise are also considered. The impact of this extra Lorentz oscillator on the peak position, width, and shape of the bump is investigated. Synthetic extinction curves are generated to model actual ones exhibiting a wide range of FUV curvatures. Physical mechanisms which might be of relevance to explain the variations of these optical properties are discussed.