Small particles tend to connect to each other and create large geometries, namely aggregates. To simplify the light scattering simulation process, they are usually modelled as assemblies of spheres positioned in point contact. This is a rough approximation because connections between them always exist. In this work we present answers to the three following questions: which optical properties of fractal-like aggregates are strongly dependent on the particle shape, what is the magnitude of the relative extinction error σCext when non-spherical particles are modelled as spheres and whether the relative extinction error σCext is dependent on the aggregate size Np. The paper was aimed at tropospheric black carbon particles and their complex refractive index m was based on the work by Chang and Charalampopoulos. The incident wavelength λ varied from λ = 300nm to λ = 900nm. For the light scattering simulations the ADDA algorithm was used. The polarizability expression was IGT_SO (approximate Integration of Greens Tensor over the dipole) and each particle, regardless of its shape, was composed of ca. Nd ≈ 1000 volume elements (dipoles). In the study, fractal-like aggregates consisted of up to Np = 300 primary particles with the volume equivalent to the volume of a sphere with the radius rp = 15nm. The fractal dimension was Df = 1:8 and the fractal prefactor was kf = 1:3. Geometries were generated with the tunable CC (Cluster-Cluster) algorithm proposed by Filippov et al. The results show that when the extinction cross section σCext is considered, the changes caused by the particle shape, which are especially visible for longer wavelengths λ cannot be neglected. The most significant difference can be observed for the regular tetrahedron. The relative extinction error σCext diminishes slightly along with the number of primary particles Np. However, even when large fractal-like aggregates are studied, it should not be considered as non-existent. On the contrary, when light scattering diagrams or the asymmetry parameter g are needed, spherical models can be used, even with relatively small fractal-like aggregates.