Training in a simulator offers potential advantages compared to training in a non-simulated environment. Generally it is cheaper, safer, there is more control over the environment, and data collection is less complicated. These potential advantages give simulators the possibility to offer effective training. This thesis is divided into three parts, aimed at realizing cost-effective driver training and driver assessment using simulators. The first part focuses on driver performance in fixed-base simulators, the second part focuses on learning to drive in a simulator and the third part evaluates a newly developed simulator for the training and assessment of racecar drivers. Valid and reliable performance measures are required to analyze driver performance. Chapter 1 evaluates a large amount of measures for the task of braking and then stopping at a stop-sign, which is a common task for research into driver performance. A computer simulation was executed and also empirical data was used to study the performance measures. The main conclusions were that reliable and valid measures for the braking task are: speed and distance to the stop-sign at the start of braking, the stopping position with respect to the stop-sign, and a measure which indicates whether or not the deceleration was constant while the vehicle was slowed down. Chapter 2 tests eight low-cost non-vestibular acceleration and speed feedback systems: a tensioning seatbelt, a vibrating steering wheel, a motion seat, screeching tire sound, auditory beeps, a vibrating seat-pan in two configurations, and a pressure seat. For five systems, which provide longitudinal acceleration feedback, the measures of Chapter 1 were used to analyze the effect of the feedback systems on driver performance during the braking task. Chapter 2 concludes that vehicle acceleration cues can be fed back to the driver without a motion platform. The system which made the largest gain in making driver performance more realistic was the tensioning seatbelt system. Chapter 3 investigates the modality with which instructions are presented in the simulator. Generally, instructions in simulators are presented verbally. The auditory modality is a logical choice because car driving can be seen as a predominantly visual task. However, beginner drivers receive a lot of verbal instructions in a limited amount of time, and therefore it is interesting to investigate the effects of presenting the route instructions in a different modality. The experiment in the driving simulator showed that both visual and visual-auditory route-instructions resulted in less turning errors than the auditory route-instructions. The visual-auditory instructions also reduced indicator reaction times. The visual-auditory instructions were preferred by people who drove faster, and people who had low self-reported driving skill. Most people preferred the visual instructions over the auditory instructions. This experiment showed that even though the visual instructions interfere with the predominantly visual driving task according to the ‘multiple research theory’ (Wickens, 1999), they did result in better driving performance. The second part of this thesis focuses on learning to drive in a simulator. First, the didactical properties of four commercially available driving simulators are analyzed. A survey shows that the intelligent tutoring systems of current driver training simulators are mostly imitating the human instructor and that the “first principles of instruction” (Merrill, 2002a) are not implemented to their full potential. Hence, there is ample room for improvement of the didactical properties by fully exploiting the many visualization, demonstration, and performance-assessment opportunities provided by modern driving simulators. Furthermore, objective performance ratings of students can be used to provide accurate and consistent feedback-on-performance, something that is not possible in real cars, but which is often essential for effective skills training. It is recommended to use empirical experimentations to improve the instructional design of simulator-based driver training for specific learning outcomes and validate the use of the first principles of instruction to facilitate learning. The following three experiments investigate whether potential advantages which are offered by simulators can be used to teach driving skills to learner drivers. Some psychological principles concerning augmented feedback are studied and used to create a new learning environment. In the first experiment, seat vibrations which reacted to the lateral position in the lane were used to teach inexperienced drivers to drive in the middle of the right lane. There were four experimental groups: (a) on-target, receiving seat vibrations when the center of the car was within 0.5 m of the lane center; (b) off-target, receiving seat vibrations when the center of the car was more than 0.5 m away from the lane center; (c) control, receiving no vibrations; and (d) realistic, receiving seat vibrations depending on engine speed. During retention, all groups were provided with the realistic vibrations. Every participant drove five 8 minute sessions: three training sessions, one retention test directly after practice, and one retention test the following day. During practice, on-target and off-target groups had better lane-keeping performance than the nonaugmented groups, but this difference diminished in the retention phase. Furthermore, during late practice and retention, the off-target group outperformed the on-target group. The conclusion of this experiment is that off-target feedback is superior to on-target feedback for learning the lane-keeping task. During the following two experiments, the difficulty of the training was varied by changing the friction coefficient of the tire on the road. The first experiment deals with a normal road-car, while the second experiment deals with a racing car. Previous research in motor learning has shown that degrading the task conditions during practice can enhance long-term retention performance. Just like in the previous experiment, the driving task was keeping the car in the center of the right lane. The inexperienced drivers were divided into three groups: low grip (LG), normal grip (NG), and high grip (HG). All groups drove six sessions: four practice sessions, an immediate retention session, and a delayed retention session the following day. The two retention sessions were driven with normal-grip tires. The results show that LG drove with lower speed than NG during practice and retention. Transferring from the last practice session to the immediate retention session, LG’s workload decreased, as measured with a secondary task, whereas HG’s workload increased. This experiment also showed that it is possible to influence self-reported confidence level, which may have potential implications for the way drivers are trained. In the second experiment in which the tire-road friction coefficient is varied during training, we are not investigating normal car driving, but racecar driving. Now the goal is not to make people drive slower, but faster instead. Three groups of inexperienced racecar drivers were trained and tested on the same simple racetrack: low grip (LG), normal grip (NG), and high grip (HG). Just like in the previous experiment, LG drove slower than the other groups during training and the first retention session. The second retention session was driven in a different car than the training and the first retention session (Formula 1 car instead of a Formula 3 car), and in this session no differences in lap time were found between the groups. LG reported a higher confidence and lower frustration than NG and HG after each of the two retention sessions. In conclusion, practicing with low grip, as compared to practicing with normal or high grip, resulted in increased confidence but slower lap times. The third part of this thesis investigates the validity and controllability of a racing simulator. A modest validation study was performed by comparing the fastest lap times of 13 racing drivers during training sessions in the simulator to the fastest lap times these same drivers did on the same track in the real world. A correlation between the lap times was found, which indicates that the simulator has some degree of predictive value for performance in the real world. A (racing) simulator can be used for controlled experiments which are difficult to perform in reality. In different racecars, we have found large differences in gain and stiffness of the brake pedal. We assume that there exists an optimal stiffness and gain of the brake pedal for racecar drivers, but this is hard to investigate in reality. The expected performance differences are small, the time it takes to adapt the brake system is lengthy, and the environmental factors, such as grip of the tires and track, vary all the time. In two independent experiments the effect of the brake pedal stiffness on lap times is investigated. The expectations were that a softer brake pedal would be better in long brake zones, and that a stiff pedal would result in faster control inputs by the driver. The conclusions of the two experiments are that racing car drivers can deal with a large range of brake pedal stiffness, that a stiff pedal results in faster control inputs, and that the simulator is a useful tool for experiments concerning the human-machine interface which are difficult to perform in reality. To get a more detailed idea about which properties of the brake pedal are important for brake force control of racecar drivers, the gain of the brake pedal is investigated further in Chapter 9. During the last experiment participants did not drive on a virtual track, but performed a one dimensional control task. The test setup was a formula racing car cockpit fitted with an isometric brake pedal, which means that the pedal does not deflect under load and the pedal force determines the output. Four control-display gains, varying from very low to very high, were compared with two target functions; a step function and a multisine function. The control-display gain had only minor effects on root mean-squared error between output value and target value but it had large effects on build-up speed, overshoot, within-participants variability, and self-reported physical load. The results confirm the hypothesis that choosing an optimum gain involves balancing stability against physical effort.