Machine learning has emerged as an apparent panacea for managing, processing and extracting knowledge from data. Its inherent ability to process vast amounts of varying types of data has enormous implications in the mining industry. Should computer processing power advance at its promised rates, machine learning has the potential to enable real time processing of multivariate data collected during drilling, to convert this information into real time geological insights, whilst simultaneously updating the resource estimation model and making real-time mining decisions. We, however, need a reality check: data and information cannot simply be thrown to machine learning. Crudely applied, machine learning simply does not always perform as expected and can even provide misleading results, particularly when combined with the mining requisite of geo-spatially reference data.So how does one select the right method for the data or circumstance and ensure outcomes are valid? When is it appropriate to reject a machine learning outcome or continue to pursue improvements in the settings?This paper brings clarity to the discussion by comparing results from several methods as applied to typical multivariate data assay data, including descriptions of analyses of the key validation processes. The intent is to help mineral industry professionals frame their questions to enable quality outcomes from machine learning. CITATION:Rondon, O, Coombes, J and Cook, A, 2019. Machine learning limiting bias and validating for value, in Proceedings Iron Ore 2019, pp 611623 (The Australasian Institute of Mining and Metallurgy: Melbourne).