Estimating kinship is fundamental for studies of evolution, conservation, and breeding. Genotyping-by-sequencing (GBS) and other restriction based genotyping methods have become widely applied in these applications in non-model organisms. However, sequencing errors, depth, and reproducibility between library preps could potentially hinder accurate genetic inferences. In this study, we tested different sets of parameters in data filtering, different reference populations and eight estimation methods to obtain a robust procedure for relatedness estimation in Scots pine (Pinus sylvestris L.). We used a seed orchard as our study system, where candidate parents are known and pedigree reconstruction can be compared with theoretical expectations. We found that relatedness estimates were lower than expected for all categories of kinship estimated if the proportion of shared SNPs was low. However, estimates reached expected values if loci showing an excess of heterozygotes were removed and genotyping error rates were considered. The genetic variance-covariance matrix (G-matrix) estimation, however, performed poorly in kinship estimation. The reduced relatedness estimates are likely due to false heterozygosity calls. We analyzed the mating structure in the seed orchard and identified a selfing rate of 3% (including crosses between clone mates) and external pollen contamination of 33.6%. Little genetic structure was observed in the sampled Scots pine natural populations, and the degree of inbreeding in the orchard seed crop is comparable to natural stands. We illustrate that under our optimized data processing procedure, relatedness, and genetic composition, including level of pollen contamination within a seed orchard crop, can be established consistently by different estimators.