The demands on the packaging and weighing industry are continually increasing and require improvements. These in the case of dynamic package weighing reflect the accuracy and speed of weighing and the handling capability of all types of articles. Furthermore, these improvements are essential for dynamic weighing if the ever tougher directives are to be obeyed. The benefits are; increased industrial profitability, automation and customer satisfaction. The conventional conveyor transport sub-system of a dynamic weighing system (checkweigher) poses a major problem in attaining high accuracy. It has become clear from the statistical, time-domain and spectral analysis of the modelling and experimental work that conventional checkweighing can never reproduce the static weighing accuracy of the weighing sensor. This is due to the low frequency noise generated by the continuously running conveyor transport. This problem will remain as long as either the conveyor transport system is used or the system and noise modelling is not precise or simple enough to implement with an advanced signal processing technique. It has been shown that dynamic accuracy (for repeated measurements) is about 5 times worse than that of the load cell under static measurements. Also, at the resonance points of the conveyor sub-system and beyond relatively low throughput rates (45 packs/min), simple signal processing (normally a fix filter with averaging) has little or no effect on improving the signal quality. In the past, the modelling has only concentrated on the weighing sensor with little or no consideration for direct and induced oscillatory noise. However, the modelling, using subsystems and classifying the noise according to speed and subsystem, has produced simulation results that match an industrial checkweigher output adequately even after replacing the sub-system type. The model algorithms are capable of system self analysis and model determination. This Is an important step if accuracy is to be increased using predictive techniques. Furthermore this modelling has been extremely useful in the simulation and analysis work. It has given new insights into solving the accuracy problem as well as produced an unexpected spin-off in another research area. The use of a robot arm has been investigated as an innovative transport method. To justify the comparison between the two transport subsystems experimentally, a scara type of robot-arm has been used in conjunction with an industrial checkweigher. Specific trajectories and special algorithms have been investigated and developed to enhance throughput rate as well as preserve high static load cell accuracy. The analysis has shown that continuous low frequency noise present on the weighing sensor as in the case of the conventional subsystem has been eliminated. The oscillatory effects of the article have also become negligible. It has been demonstrated that the robot arm is capable of handling both a solid and loose type of article without significant degradation in accuracy. The experimental results for the robot arm have far exceeded the expectations. It has been shown that the accuracy drops from 98% to 97% with the robot-arm but from 98% to 88% with the conveyor subsystem; 98% represents the static load cell accuracy. The accuracy was measured over the throughput range 0 to about 40 packs/min. It drops with the robot-arm therefore only about 1/10 that of the conveyor sub-system. In addition, the throughput rates obtained showed that it is possible to achieve over 2 times the throughput rate change with speed. This indirectly relates to accuracy because lower speeds (example 5 m/min) produce lower disturbances even though the throughput rate (example 20 packs/min instead of 10 with the conveyor type) is higher. Hence the robot based system is superior compared to the conventional system in terms of producing higher accuracy over a reasonable throughput range. The simulation of digital adaptive FIR filters, examined with an industrial checkweigher system, has shown to maintain accuracy and increase throughput rate up to 80% over a range of speeds up to 100 m/min. The adaptive concept, to establish the optimal filter parameters and type, is based on a fast 3-D search method using successive approximation. The adaptive filtering will be implemented in the checkweigher with slight refinement to speed up the simulation algorithms used. To obtain even higher improvements, special robot designs and dynamic weighing sensors would have to be investigated in the future. In addition, low frequency signal processing has to be considered using predictive signal processing methods.