The rolled leaf trait, long considered to be a key component of plant architecture, represents an important target trait for improving plant architecture at the population level. We therefore performed linkage mapping using a set of 262 highly variable RILs from two rice cultivars (Minghui 63 and 02428) with minor differences in leaf rolling index (LRI) in conjunction with GWAS mapping of a random subset of the 1127 germplasms from the 3K Rice Genomes Project (3K Rice). A total of seven main-effect loci were found to underlie the transgressive segregation of progenies from parents with minor differences in LRI. Five of these loci were previously identified and two (qRl7b and qRl9b) are newly reported with additional evidence from GWAS mapping for qRl7b. A total of 18 QTLs were identified by GWAS, including four newly identified QTLs. Six QTLs were confirmed by linkage mapping with the above RIL population, and 83.3% were found to be consistent with previously reported loci based on comparative mapping. We also performed allele mining with representative SNPs and identified the elite germplasms for the improvement of rolled leaf trait. Most favorable alleles at the detected loci were contributed by various 3K Rice germplasms. By a re-scanning of the candidate region with more saturated SNP markers, we dissected the region harboring gRl4-2 into three subregions, in which the average effect on LRI was 3.5% with a range from 2.4 to 4.1% in the third subregion, suggesting the presence of a new locus or loci within this region. The representative SNPs for favorable alleles in the reliable QTLs which were consistently identified in both bi-parental mapping and GWAS, such as qRl4, qRl5, qRl6, qRl7a, and qRl7b will be useful for future molecular breeding programs for ideal plant type in rice.