Since 1972, the International Crops Research Institute for the Semi‐Arid Tropics (ICRISAT) has maintained a large collection of sorghum in India. The collection size has continuously increased, and the total number of accessions at present conserved in the gene bank has reached about 36 000 accessions. The need to help management was considered, and this study was conducted to establish core collections. This sorghum collection was earlier stratified into four clusters according to the photoperiod sensitivity. Then, considering the core collection strategy, we used three random sampling procedures to determine the specific accessions to be included in the core [i.e., a constant portion (Core C), a proportional (Core P), and a proportional to the logarithm (Core L)] of the photoperiod group size sampling strategy. Both the Core C and L were significantly different from the landrace collection with better representation of the smallest groups, such as landraces insensitive to photoperiod. Despite differences between the three core collections, estimates of global diversity through the Shannon‐Weaver Diversity Indices were of the same magnitude as the landrace collection. When compared, the Core C and L were significantly different. Core L sampled better for the characters, the race, and the latitudinal classes that were related to the photoperiod‐sensitive landraces. Thus, for establishing a core collection with the widest range of adaptation to photoperiod, we propose the use of a logarithmic sampling strategy, which identifies a broadly adapted set of genotypes.