As the number of traortation goods as well as traortation cost has increased, the owners of the industry are motivated to seek for methods that reduce the costs and subsequently, increase the profit. Therefore, vehicle routing and loading problems have become two major issues for reducing traortation costs recently. one-to-one pickup and delivery problem is one among the pickup and delivery problem groups, in which goods are traorted between pickup and delivery nodes by available vehicles. For the first time, one-to-one pickup and delivery problem with three dimensional loading constraints and time window are considered in this thesis for hetergenous fleet of vehicles. To this end, an adaptive large neighbourhood search algorithm is developed for the routing problem. This algorithm achieves acceptble results for small size problems and it could also yield optimal objective for all problems. In addition, this algorithm showed also a good performance for large size problems, improving 76% of benchmark instances up to 2.66%. Also, we formulated two mathematical models for the integrated pikcup and delivery along with loading problems based on two approaches. In the first approach, stage indices were used as variables to integrate pickup and delivry problem with loading ones. In the second approach, a binery variable is used that defines the stage of visiting a node. Using this binery variable, the number of variables order is decreased. The two proposed models are assessed and it is revealed that the model using the second approach could solve larger size problems faster comparing with the model using the first approach. In both models, LIFO, stability and fragility constraints are formulated differently than the existing formulation. A heuristic based on the extreme points is used to solve the loading problem. To improving the performance of heuristic, order of boxes of each customer varied, in case the loading of boxes become infeasible during the first iteration. The number of order variations of each customer depends on the number of boxes. The proposed method could search more loading patterns and consequently, the quality of heuristic is increased. The loading heuristic is then integrated within the routing algorithm and the acquired results for small size problems were compared with mathematical model results. Intergrated algorithm yielded the optimal objective in 99% of the problems without stability and fragility constrains. Also, in case of considering all the constraints, the mathematical model without stability and fragility constraints is used as the lower bound. In this case, integrated algorithm achieved 0.11% difference with the optimal value. Comparing the acquired results with mathematical model results approved high performance of the inegrated algorithm. Finally, the acquired results for benchmark instances in large sizes are reported and the sensivity of each loading constarint to the objective is examined. Experiments showed that LIFO constraint has the highest infulence on objective increasing it by 16%.