We report a modelling study to investigate the effects of constraining the inversion of Electrical Resistivity Tomography (ERT) data, from surface arrays, with reference models derived from supplementary resistivity data such as borehole resistivity logs, resistivity cone penetrometry (RCPT), and electromagnetic survey. A synthetic resistivity site model of a highly resistive (200 Ωm) sand and gravel lens in a low resistivity (30 Ωm) clay till was constructed to test the approach. Synthetic Wenner ERT field data were generated from the synthetic site model and contaminated with fifty sets of Gaussian noise with standard deviation of 2%, and a further fifty sets of Gaussian noise with a standard deviation of 5% of the measurement value. Five structured reference models were constructed incorporating top and basal boundaries of the high resistivity lens, simulating one hit with targeted RCPT, while varying lens width. The noisy ERT data were inverted with 1. an homogenous reference model (blind inversion), and 2. the structured reference models (guided inversion).The results show that, for blind inversions, the resistivity of a small lens with a resistivity of 200 Ωm will be typically underestimated by about 100 Ωm, which is half its value, in the presence of Gaussian noise; this is a consequence of equivalence. Better reconstructions can be achieved using structured reference models, provided that these are structurally similar to the synthetic site model representing the true geoelectrical structure. More importantly, the reference models and resulting solution models that are close approximations to the actual subsurface structure can be identified without knowledge of the synthetic site model. This is done by comparing the misfits between the solution and reference models, which act as a proxy for the misfit between the solution models and the synthetic model, even when the data are noise-contaminated. Essentially, the approach developed uses the additional (non-ERT) data to identify a limited set of the possible solutions to the blind inversion, which are also compatible with that additional data.ii