In this thesis, we design and analyze autonomous vehicle storage and retrieval systems (AVS/RSs) via queuing network and simulation models. An AVS/RS is an automated unit-load (UL) storage and retrieval system based on autonomous vehicle (AV) technology. It represents a relatively new technology for automated, UL storage systems (Malmborg, 2002). AVs function as storage/retrieval (S/R) devices. A key distinction of AVS/RSs relative to traditional crane-based automated storage and retrieval systems (AS/RS) is the movement pattern of the S/R device. In AS/RSs, ULs are stored or retrieved by aisle-captive storage cranes capable of simultaneous movement in the horizontal and vertical dimensions. The vehicles in an AVS/RS share a fixed number of lifts for vertical movement and follow rectilinear flow patterns for horizontal travel. To be able to benefit from both modeling approaches, we develop simulation and analytical models for a particular AVS/RS in this thesis. The simulation modeling approach simulates the sequence of events that could occur over a period time via a computer program. There are some advantages and disadvantages with the simulation methodology. The most important advantage is the ability to model complex systems in great detail, so it provides more accurate results. For example, a verified and validated simulation model could provide estimates of key performance measure that are very close to those seen in the actual system. However, this high accuracy comes at the expense of high modeling and computational effort. Developing a detailed, more accurate simulation model for a large system is time consuming. Analytical modeling uses mathematical relationships between inputs and outputs. The most important advantage of an analytical method is that it is typically not time consuming. It can evaluate the system's performance in a reasonable time. However, to develop an analytical model for a complicated system is not a simple task. Also, changing an assumption in an analytical model may render the model invalid. These are some of the disadvantages of analytical modeling approach. When properly designed, analytical models, however, are capable of providing reasonably accurate estimates of complex systems in a relatively short time. The analytical model of the system we study is modeled as a semi-open queuing network (SOQN) model. An SOQN consists of jobs, pallets and servers. Each job is paired with a pallet and the two visit the set of servers required for processing the job in the specified sequence. In the AVS/RS, we assume the storage/retrieval (S/R) transactions are the 'jobs' and the AVs are the 'pallets'. If an S/R transaction requires a vertical movement, it uses a lift. The lifts and horizontal travel times to and from a storage space are modeled as servers. We solve the SOQN of the AVS/RS by an approximate analytical model and the matrix geometric method (MGM). We use the simulation model of AVS/RS to validate the analytical models. Thus, we compare the approximate, MGM and simulation results. As a result for the problems we tested, MGM estimates a key performance measure--waiting time for storage and retrieval transactions in the external queue--better than the approximate analytical results. We also perform a simulation based experimental design to identify factors affecting the performance of AVS/RS. In addition, we study the rack configuration design of the AVS/RS warehouse using a simulation based regression analysis. In the design of experiment (DOE), we consider three performance measures, namely average transaction cycle times, lift utilizations and vehicle utilization, as well as four factors--vehicle dwell point policy, scheduling rule, location of input/output (I/O) point on the ground floor, and interleaving policy. The DOE results show that all the pre-defined factors are significant on the performance measures. After DOE, we implement Tukey's test to find out the best experiment(s) that reflect the performance measure. The experiment in which vehicles dwell near the lift location, shortest distance traveled scheduling rule, locating the I/O point in the middle of the x-axis of the warehouse, and opportunistic interleaving policy is determined to be the best experiment. In the regression analysis, we develop thirty regression functions based on various number of vehicles and lifts, and arrival rate scenarios in the system. The regression functions are developed in terms of number of tiers, aisles and bays input variables. We consider five performance measures. After obtaining the regression functions, we optimize them using the LINGO software. In many cases, the results suggest that the warehouse design be as long as is practically possible in the axis. Much of the data for the analytical and simulation models come from a warehouse in France that uses the AVS/RS.