Load models play an essential role in many studies, including calculating voltage drops and technical losses in distribution systems, for distributed generator (DG) integration planning, and in tariff analysis and design models. The Herman-Beta transform used in the low voltage network modelling studies in South Africa is based on loads modelled as Beta probability density functions. Recently, the transform was extended to make it useful also for probabilistic load flow modelling in medium voltage (MV) networks with non-unity power factor loads and DGs. The electricity supply industry in South Africa has transformed and saw an increased penetration of Independent Power Producers as a result of the government encouraged the renewable independent power procurement programme (REIPPP). There has also been a steady decrease in the costs of procuring power from renewable energy sources, mainly from photovoltaic (PV) systems. South Africa also saw significant tariff increases in the recent past. These have resulted in both new load patterns and uncertainties in the power systems inputs required for network planning and tariff development. Other factors affecting loads and renewable energy output include weather, location and economic factors. Load models are essential for technical and tariff studies. Long term and short term planning models in both technical and tariff modelling require information about the usage behaviour of customers. Planning cannot be separated from the financial impact and tariffs in general. The literature review indicated that planning has the objective of designing a network for optimal usage, thus minimising the costs and deferring investment where possible. Load patterns have been recognised to represent the usage behaviours of customers better and these behaviours influence the planning parameters. There have been studies by numerous researchers to extract parameters from the load profiles for load flow modelling and simulation purposes. The same challenge exists for South Africa, where there has been progress made on the development of LV models, and the same is not replicated in the MV network space. The derivation of load models primarily involves the classification of loads, identifying and estimating the parameters of loads, and assigning load profiles to different loads for studies. Customer measurements are an essential input in load model development and load estimation. Identification of parameters is one of the areas where research is ongoing since there is no global consensus on which attributes best describe customer load profiles. In this study, a proposition on how the parameters for technical and tariff analysis models should be defined was made. The use of 24-hour load profiles to classify calendar days into typical days was also suggested. The availability of measurements data made it possible to develop load models for MV and conduct a study on actual customer data. The customers' measurements data, made it possible to identify the parameters and develop load models that could be used for technical and tariff analysis and conduct a pilot study to evaluate the load models. This study proposes a load model that can be used to model typical days and to model customer loads. The load models proposed here uses the k-means clustering algorithm as the basis for classification. The load models enable the classification of loads and assignment of load profiles accordingly. The results of this study indicated that load parameter models could be extracted from the customer measurements, for technical and tariff studies in distribution networks. It has also been possible to identify and determine the parameters from the load profiles and proposed a process for developing a load model for technical, economic and tariff analysis. The results also indicate that of the five identified parameters, the most significant parameters that affected the clustering results were the load factor, average power and the normalised peak usage parameter when the results of each of the factors were compared on an individual basis. The study also revealed improvements to the clustering results when all the parameters identified in this study were combined and a PCAbased clustering algorithm was used. Finally, the results indicate that the loads in the different economic activitybased classifications do not necessarily have similar shapes although they belong to the same cluster. The modelling process developed in this study may be implemented by utilities for determining load parameter models for MV feeders when measurements are available. The process may also be used to guide future data collection.