The use of appropriate statistical methods has a key role in improving the accuracy of selection decisions in a plant breeding program. This is particularly important in the early stages of testing in which selections are based on data from a limited number of field trials that include large numbers of breeding lines with minimal replication. The method of analysis currently recommended for early-stage trials in Australia involves a linear mixed model that includes genetic relatedness via ancestral information: non-genetic effects that reflect the experimental design and a residual model that accommodates spatial dependence. Such analyses have been widely accepted as they have been found to produce accurate predictions of both additive and total genetic effects, the latter providing the basis for selection decisions. In this paper, we present the results of a case study of 34 early-stage trials to demonstrate this type of analysis and to reinforce the importance of including information on genetic relatedness. In addition to the application of a superior method of analysis, it is also critical to ensure the use of sound experimental designs. Recently, model-based designs have become popular in Australian plant breeding programs. Within this paradigm, the design search would ideally be based on a linear mixed model that matches, as closely as possible, the model used for analysis. Therefore, in this paper, we propose the use of models for design generation that include information on genetic relatedness and also include non-genetic and residual models based on the analysis of historic data for individual breeding programs. At present, the most commonly used design generation model omits genetic relatedness information and uses non-genetic and residual models that are supplied as default models in the associated software packages. The major reasons for this are that preexisting software is unacceptably slow for designs incorporating genetic relatedness and the accuracy gains resulting from the use of genetic relatedness have not been quantified. Both of these issues are addressed in the current paper. An updating scheme for calculating the optimality criterion in the design search is presented and is shown to afford prodigious computational savings. An in silico study that compares three types of design function across a range of ancillary treatments shows the gains in accuracy for the prediction of total genetic effects (and thence selection) achieved from model-based designs using genetic relatedness and program specific non-genetic and residual models. Supplementary materials accompanying this paper appear online.