Comprehending the properties of complex nanoscale materials requires not just study of their morphology, but also determining the distribution and quantity of specific elements or phases, and the nature of bonding within these. Here we present quantitative 3D elemental and bonding mapping of a complex boron nitride based nanoparticle. This is achieved through a combination of electron energy‐loss spectroscopy (EELS) in the scanning transmission electron microscope (STEM), novel EELS analysis methods and compressed sensing tomographic reconstruction [1]. For this study, a low‐loss and core‐loss STEM‐EELS tilt series of spectrum images was recorded over the angular range −70° to 70° with a 17.5° tilt increment. The experiment was performed at 80 kV using a Tecnai Osiris with Gatan Enfinium spectrometer equipped with DualEELS. Figure 1a shows the high angle annular‐dark field (HAADF) tilt‐series images of the nanoparticle studied. The nanoparticle clearly possesses intricate structure, but the HAADF images are not fully revealing, obviating need for tomographic and analytical investigation. EELS can achieve accurate absolute quantification of elemental composition without the need for standards, opening the door to quantitative analytical tomography. Moreover, the fine structure exhibited within the first tens of eV above an EELS ionisation edges is related to the local density of states, and hence, carries a wealth of information about the electronic environment of the ionised atom. However, direct measurement of the fine structure of pure compounds is only possible in homogeneous materials and in atomically resolved EELS of two‐dimensional mono‐layered materials. More commonly, EELS measurements comprise a linear combination of the fine structure corresponding to different atomic environments. Here we have devised a novel method to extract the fine structure of individual compounds from a multi‐dimensional EELS dataset, based on a combination of curve fitting [2] and blind source separation [3]. Major practical complications with curve fitting for EELS quantification, especially in multi‐dimensional datasets, are ill‐conditioning and divergence of non‐linear optimisation. To address this we have developed a new parallel Smart Adaptive Multidimensional Fitting (SAMFire) algorithm that learns the starting parameters from the dataset as the fitting progresses [4]. The analysis reveals that the particle is composed of boron (in different compounds), nitrogen, oxygen, carbon, silicon and calcium (Figure 1b, c). The EELS data analysis was performed using HyperSpy [5]. Tomographic reconstruction of the obtained tilt series of EELS elemental and bonding maps was performed using the FISTA algorithm with 3D total variation regularization [6]. Figure 2 displays a 3D visualization of the three boron compounds found in the sample, namely boron oxide, pure boron and boron nitride. Despite the small number of tilt series maps, the tomographic reconstruction reveals comprehensively the details of this complex 3D structure and provides new insight on the growth mechanism of the particle.