The aim of this study is to estimate the uncertainty of a portable X-ray fluorescence (p-XRF) instrument for the (semi-quantitative) analyses of tiles with underglaze decoration. Before starting the campaign of on-site measurements, the optimum acquisition time and the most accurate calibration mode were selected. For this purpose, the elemental composition of two glass standards of NIST (SRM610 and SRM612) and a Corning A standard were measured with varied times (5–360 s) and in different calibration modes (Mining, Mining Light Elements, Soil, and Rare Earth Elements). Afterwards, a set of blue-and-white tiles that was unearthed at Iznik Tile Kilns Excavation between the dig seasons of 2015 and 2019 was examined with p-XRF by selecting ten points of measure from each layer (body, transparent glaze, and blue coloured areas). The elemental composition of different layers was evaluated by means of the intragroup and intergroup data. They were also compared to the previous studies and found that the corrosion-free, homogeneous, and non-porous surfaces decrease the relative standard deviation (RSD) by increasing the consistency of the compositional data. The major elements found in the matrix of each layer (Al and Si for the body, Pb and Sn for the glaze) have the lowest value of RSD, as expected. However, the comparison of the data with the analysis of the reference materials showed that the content of Mg and also Si, which belong to the low-Z elements group, is shifted relatively towards the higher compositional values. The impossibility of measuring the elemental composition of sodium does not hinder the classification of the samples. Although the transition metals have very low concentrations, p-XRF measurements appear rather consistent and the intrinsic scattering of the data observed for a single artefact is largely smaller than those observed for the tiles of different historical buildings. Thus, it allows the classification to be made related to the different techniques used.