The Bovine Long Read Consortium aims to use long read sequencing of cattle at population scale to characterise structural variation of the genome. This pilot study used sequences from 41 animals from two breeds, including two trios at 45X. The trios were scaled back to lower read depths. SV recall and precision was lower at 10-15X versus 45X, however this may be an acceptable compromise for population scale studies to spread costs over large numbers of animals. However, if the purpose is to discover a deleterious mendelian mutation, closer to 45X is preferred. SNP and INDEL were called from various read depths and compared to a filtered set of high-quality calls from short read sequences. At each read depth the recall and precision of SNP was considerably higher than INDEL. At 10X, SNP recall was 0.86 and 0.99 at 45X. The precision for SNP and particularly INDEL suggested that long-read variant calls include a relatively high, but likely overestimated proportion of false positives. After merging and joint SV calling across all animals, levels of sporadic missing genotypes was less than 5% per animal (most 10-25X). A total of 76,572 SVs were detected across all samples, one third of which were segregating in one breed. Using variant effect predictor annotations, a much higher proportion of SV compared to SNP had high impact effects. Insertions and deletions tended to be smaller and duplications larger. Insertions and deletions more often segregated across both breeds, while inversions were more often breed specific. Few duplications were detected but they were slightly more likely to be breed specific. The results highlight that it would be beneficial to build a large reference population to understand the structural variation, to impute these SV into large phenotyped populations and to explore the impact of SVs on traits of interest.