ABSTRACT We present a non-parametric reconstruction of the rotation curves (RCs) for 88 spiral galaxies using the LOESS (locally weighted scatterplot smoothing) + SIMEX (simulation and extrapolation) technique. In order to compare methods, we also use a parametric approach, assuming core and cuspy dark matter (DM) profiles: pseudo-isothermal (PISO), Navarro−Frenk–White (NFW), Burkert, Spano, the soliton, and two fuzzy soliton + NFW. As a result of these two approaches, a comparison of the RCs obtained is carried out by computing the distance between the central curves and the distance between the 1σ error bands. Furthermore, we perform a model selection according to two statistical criteria, the Bayesian information criterion and the value of $\chi ^2_{\rm red}$. We work with two groups. The first is a comparison between PISO, NFW, Spano and Burkert, showing that Spano is the most favoured model satisfying our selection criteria. For the second group, we select the soliton, NFW and fuzzy models, resulting in soliton as the best model. Moreover, according to the statistical tools and non-parametric reconstruction, we are able to classify galaxies as core or cuspy. Finally, using a Markov chain Monte Carlo method, for each of the DM models we compute the characteristic surface density, μDM = ρsrs, and the mass within 300 pc. We find that there is a common mass for spiral galaxies of the order of 107 M⊙, which is in agreement with results for dSph Milky Way satellites, independent of the model. This result is also consistent with our finding that there is a constant characteristic volume density of haloes. Finally, we also find that μDM is not constant, which is in tension with the literature.