Seismic tomography has seen a significant improvement since the end of the twentieth century with the expansion of seismic station networks and theoretical and computational advances leading to more and more detailed models of the Earth’s deep interior. While features such as subduction zones are relatively easy to image, other features such as mantle plumes continue to be difficult to resolve. Furthermore, differences in datasets, theory and methods used to build tomographic models result in different features appearing inconsistently, and a lack of uncertainty quantification impedes distinguishing features from noise. This work uses a spherical wavelet transform and random noise realisations to objectively quantify the probability of plume-like features in six recent global tomographic models. We find only two features, in the Pacific and East Africa, that consistently and confidently appear in at least five models. Hawaii and Iceland plumes are found in only up to two models, as are plumes around the Southwest and Southeast Indian ridges. From this we conclude that great care must be taken when visually interpreting plume-like features in tomographic models, as few can be clearly distinguished from noise or small-scale artefacts. Additionally, high correlations are found between plume probability maps and shear wave speed within the boundary of large low shear velocity provinces (LLSVPs) at the core-mantle-boundary in all models, reinforcing the notion that some plumes may be rooted at the LLSVPs. While the focus of this work is on mantle plumes, the tool developed to examine tomography models is flexible and can be used to assess other features of the tomographic models.