Most planets currently amenable to transit spectroscopy are close enough to their host star to exhibit a relatively strong day to night temperature gradient. For hot planets, this leads to cause a chemical composition dichotomy between the two hemispheres. In the extreme case of ultra hot jupiters, some species, such as molecular hydrogen and water, are strongly dissociated on the day-side while others, such as carbon monoxide, are not. However, most current retrieval algorithm rely on 1D forward models that are unable to model this effect. We thus investigate how the 3D structure of the atmosphere biases the abundances retrieved using commonly used algorithms. We study the case of Wasp-121b as a prototypical ultra hot Jupiter. We use the simulations of this planet performed with the Substellar and Planetary Atmospheric Radiation and Circulation (SPARC/MIT) global climate model (GCM) and generate transmission spectra that fully account for the 3D structure of the atmosphere with Pytmosph3R . These spectra are then analyzed using the TauREx retrieval code. We find that such ultra hot jupiter's transmission spectra exhibit muted H 2 O features that originate in the night-side where the temperature, hence the scale-height, is smaller than on the day-side. However, the spectral features of molecules present on the day-side are boosted by both its high temperature and low mean molecular weight. As a result, the retrieved parameters are strongly biased compared to the ground truth. In particular the [CO]/[H 2 O] is overestimated by one to three orders of magnitude. This must be kept in mind when using such retrieval analysis to infer the C/O ratio of a planet's atmosphere. We also discuss whether indicators can allow us to infer the 3D structure of an observed atmosphere. Finally we show that Wide Field Camera 3 from Hubble Space Telescope (HST/WFC3) transmission data of Wasp-121b are compatible with the day-night thermal and compositional dichotomy predicted by models. Caldas et al. (2019) highlighted systematic biases on retrieved temperatures using a 1D retrieval model TauREx (Waldmann et al. 2015a,b). However, this earlier study only looked at atmospheres with a homogeneous composition to focus on ther-Article number, page 1 of 13 arXiv:2003.05943v1 [astro-ph.EP] 12 Mar 2020 mal effects. To that purpose, we focus on Wasp-121b (Evans et al. 2016Parmentier et al. 2018) as a prototype for UHJs. The complexity of the UHJ atmospheres describes in our GCM model may suggest that we would need a more complex framework to analyze the transmission spectra of those particular planets. Thus, we will need fully 3D forward models to simulate realistic transmission spectra to better fit transit observations of UHJs. Hence, 1D retrieval models such as TauREx are probably biased in their analysis.Hereafter, we first describe the observational chain used to simulate JWST observations in Sect. 2. Then, in Sect. 3, we explain the numerical experiments done to unravel the biases starting from a very simple, parametric modeling of Wasp...