Geological domains describe the mineralisation state and stratigraphic location of different geological regions of mines. They play a very important role at the geological modelling stage as weight percentage of chemical elements information within each geological domain is assumed to be correlated while information coming from different geological domains is considered uncorrelated. Having correct domain assignments to both the chemical assays and the 3D space of modelling is critical for building a reliable geological model of the mine.As standard practice significant effort is invested into the interpretation of the deposit and identification of the geological domains for the chemical assays and the 3D geological space. Despite these efforts, there are some chemical assays with ambiguous geological domain where the exploration assay domain and the block domain do not match. Incorrectly including a sample in the wrong domain can cause a large change in the predicted weight percentage of chemical elements estimation for that block. Therefore, the major challenge of grouping the spatially correlating weight percentage of chemical elements is how to determine which of these domains best fits these misclassified samples.This work proposes machine learning based autonomous technique for identifying to which geological domain (the exploration assay domain or the block domain) the misclassified sample belongs. The proposed methodology is based on Gaussian Process (GP) classification. The GP machine learning approach consists of training and inference stages. At the training stage the GP automatically learns the properties of different geological domains using the chemical assays with unambiguous domain assignment. Then, at the inference stage the GP makes predictions for the geological domains of the assays with ambiguous initial domain assignment.We tested the proposed technique by conducting a case study for an iron ore deposit located in Western Australia. The results demonstrate that the application of the proposed GP technique removes statistically significant deviations in the assay values in many regions: mineralised ore region, unmineralised region, highly weathered region and intrusive material that is highly variable. In this case study the proposed methodology autonomously identified the geological domains for more than 50 per cent of samples which are initially ambiguously classified. For the majority of the assays the GP predicted domain agrees with the geological domain assigned by the exploration assay.CITATION:Balamurali, M and Melkumyan, A, 2014. Automated identification of geological domains for exploration assays with ambiguous initial domain assignment in an iron ore deposit, in Proceedings Ninth International Mining Geology Conference 2014 , pp 99-106 (The Australasian Institute of Mining and Metallurgy: Melbourne).