The homogeneous and stable dispersion of carbon nanotubes (CNTs) in solvents is often a prerequisite for their use in advanced materials. Dispersion procedures, reagent concentration as well as the interactions among reagent, defective CNTs and near-perfect CNTs will affect the resulting CNT dispersion properties. This study, for the first time, presents a detailed comparison between two different approaches for dispersing CNTs. The results enhance our understanding of the interactions between surfactant, defective CNTs and near-perfect CNTs and thus provide insight into the mechanism of CNT dispersion. Dispersions of “as-produced” short multi-walled carbon nanotubes (MWCNTs) in N,N-dimethylformamide were prepared by two different surfactant (Triton X-100) assisted methods: ultrasonication and ultrasonication followed by centrifugation, decanting the supernatant and redispersing the precipitate. Visual observation and UV–visible spectroscopy results showed that the latter method produce a more stable dispersion with higher MWCNT content compared to dispersions produced by ultrasonication alone. Transmission electron microscopy and Raman spectroscopic investigations revealed that the centrifugation/decanting step removed highly defective nanotubes, amorphous carbon and excess surfactant from the readily re-dispersible near-perfect CNT precipitate. This is contrary to other published findings where the dispersed MWCNTs were found in the supernatant. Thermogravimetric analysis showed that 95 % of Triton X-100 was removed by centrifugation/decanting step, and the remainder of the Triton X-100 molecules is likely randomly adsorbed onto the MWCNT surface. Infrared spectral analysis suggests that the methylene groups of the polyoxyethylene (aliphatic ether) chains of the residual Triton X-100 molecules are interacting with the MWCNTs.