Time-lapse or 4D Seismic data (4DS) are important constraints in reservoir studies because they enable monitoring of saturation and pressure changes that result from hydrocarbon production. 4DS have been quantitatively added, along with production data, in history matching or data assimilation (DA) procedures to reduce uncertainty and improve production forecasts. Before performing quantitative studies, it is important to ensure that the 4DS are reliable, with minimal artifacts such as side-lobe effects, that can disturb the identification of anomalies. In this work, we propose different ways of treating 4DS in DA for a real reservoir. Explicitly, we evaluated the impact on DA results when considering different amounts of 4D information and three treatments to the identified artifacts. The treatments were: ignoring them, excluding them from DA, or defining no seismic changes at their locations. The results show that well and seismic matches are improved when 4DS are assimilated, also improving the production predictions. Despite being a thin reservoir, assimilating two single-layer maps allowed us to predict relevant observed dynamic behaviour, such as the evolved gas trapped in the lower interval. Furthermore, when a treatment was applied to the artifacts, they produced better models than using a single two-layer map (with lower production errors and visually closer impedances to the observed data). Our recommendation is the assimilation of well and 4DS data, with the exclusion of unreliable information, for better life-cycle decisions.