Many different types of multiparental populations have recently been produced to increase genetic diversity and resolutionin QTL mapping. Low-coverage, genotyping-by-sequencing (GBS) technology has become a cost-effective tool in these populations,despite large amounts of missing data in offspring and founders. In this work, we present a general statistical framework for genotypeimputation in such experimental crosses from low-coverage GBS data. Generalizing a previously developed hidden Markov model forcalculating ancestral origins of offspring DNA, we present an imputation algorithm that does not require parental data and that isapplicable to bi- and multiparental populations. Our imputation algorithm allows heterozygosity of parents and offspring as well aserror correction in observed genotypes. Further, our approach can combine imputation and genotype calling from sequencing reads,and it also applies to called genotypes from SNP array data. We evaluate our imputation algorithm by simulated and real data sets infour different types of populations: the F2, the advanced intercross recombinant inbred lines, the multiparent advanced generationintercross, and the cross-pollinated population. Because our approach uses marker data and population design information efficiently,the comparisons with previous approaches show that our imputation is accurate at even very low (,13) sequencing depth, inaddition to having accurate genotype phasing and error detection.