Context Research works on operator monitoring and affective computing underline the benefit of gathering several signal modalities to improve accuracy for an objective mental state diagnosis (Rahman, Begum and Ahmed, 2015). The cardiovascular activity is one of the most utilized systemic measures to assess Cognitive Workload (CW) and emotion. However, the respiration contribution is usually neglected or employed mainly for artifact rejection and scarcely utilized for CW estimation in ecological situations (Grassmann, Vlemincx, von Leupoldt, Mittelstäd and Van den Bergh, 2016). On the other hand, from an ergonomic standpoint, the inclusion of new sensors to collect physiological data could be difficult to ensure high quality signals without hindering operator’s comfort, this is the case when driving smart vehicles. Objectives The purpose of this study is twofold. Firstly, it aims at verifying the advantage of incorporating Respiratory Rate (RR) as feature, with regard to Heart Rate (HR), so as to evaluate driver’s activity and CW variations. Secondly, this study aims at checking the feasibility and accuracy to extract RR from the ECG recordings in order to save on sensors. Method Eighteen healthy subjects (10 males, 22.7±1.4 years) participated in the study. None of them had a history of cardiorespiratory diseases. A valid driving license for at least 3 years was required. The participants performed two different cognitive tasks, 5-minutes length approximately, requiring different CW demands: a beep counting for the Low Cognitive Workload (LCW) condition and a mental displacement within a previously memorized 5× 5 numerical grid including an arithmetic task for the High Cognitive Workload (HCW). The participants carried out these cognitive tasks during two activities: a Single Task (ST) as well as simultaneously to a Driving Task (DT) in a simulator. They had to drive in an urban residential zone with sparse traffic. The ECG and respiration signals were recorded by MP150 Biopac system. The ECG-derived respiratory signal was extracted by using Matlab R2017b according to the algorithm implemented by Moody, Mark, Zoccola and Mantero (1995), 4.5 minutes length segments were selected for feature extraction. An analysis of variance (ANOVA) of two factors, the first one representing the activity (ST and DT) and the second one referring to the CW demand (LCW and HCW) was performed for the statistical contrasts. Tukey correction was applied for post hoc analysis. A paired t-test and Pearson’s correlation were computed to estimate ECG-derived RR accuracy. All the statistical analyses were carried out by using SPSS 13.0 software. Results Concerning HR, a main effect of activity (F (1, 11)= 21.2; p<. 01; ɳ2=. 658) and CW demand (F (1, 11)= 32.68; p<. 01; ɳ2=. 748) were found, showing an increase in HR while driving in comparison to ST and for HCW. Their interaction was not significant. Similarly to HR, RR increased while driving (F (1, 14)= 48.19; p<. 01; ɳ2=. 775), but no effect of CW demand was found. Conversely to HR, an interaction between the activity and CW demand was found for RR (F (1, 14)= 11.55; p<. 01; ɳ2=. 452). Whereas the post hoc analysis showed that RR increased for HCW in comparison to LCW under ST condition (p<. 01), no significant differences between LCW and HCW were found while driving. The correlation between HR and RR was r= 0.44. Regarding the estimation of the RR via ECG by comparing the outcomes with the RR from the original respiratory signals, a significant correlation of r=. 86 (p<. 01) was achieved. Even if an overestimation of 1.63 inspirations/min on average was evidenced (p<. 01), the inaccuracy of …