The use of a spectrum is ubiquitous across chemistry as it is used to correlate chemical structure to an interpretable signal. On the other hand, molecular surfaces are a common representation of a molecule as certain chemical concepts are thought to occur on the boundaries of a molecule. Motivated by signal processing and computational geometry, we represent a function on the molecular surface as a spectrum that encodes information about the geometry of the surface and its properties. This poster shares our ideas and results for developing alignment-free and property-specific molecular similarity measures for predicting molecular properties in cheminformatics.