The fraction of relatives in large genetic databases is continuously increasing, making it more vital to uncover the relationships among samples for further downstream analysis. Relatedness inference is a fundamental step for genetic association studies, population genetics, and genealogy. For close relatives, it is even possible to infer specific relationship types. However, due to the randomness of recombination and inheritance, different pairs of relatives, even in different degrees of relatedness, can have similar amounts of identical by descent (IBD) sharing. This makes relatedness inference difficult and especially causes ambiguities for inferring relationship types of the same degree. We first present an analysis that explores the possibility of improving the accuracy of relatedness inference by adding IBD segment numbers. We investigated the importance of IBD segment numbers between the pair of relatives via both a theoretical information theory analysis and a machine learning classification approach. Our study showed that the IBD segment number adds information for the relatedness inference in general, but the improvement of accuracy is weakened by the IBD detection error in practice. Next, we describe CREST, an accurate and fast method to identify specific relationship types of close relatives using multiway IBD sharing. More specifically, for a given second degree relative pair, we leveraged their mutual relatives to determine their relationship types---grandparent grandchild (GP), avuncular (AV), and half siblings (HS). CREST achieved high sensitivities when tested in both simulated and real dataset with sufficient mutual relatives. CREST also identifies the directionality for parent child (PC), AV, and GP pairs and has the potential to be extended to identify more distant relatives. Lastly, with the aid of IBD segments from relatives, we developed HAPI-RECAP to reconstruct parental genome from a set of genotyped siblings and their relatives using a combination of family-based phasing and IBD sharing. For families with eight or more children, HAPI-RECAP can reconstruct most of genotypes for two parents, with Comparable error rates to direct genotyping, using only genotype data from children. For smaller families with four to seven children, HAPI-RECAP is able to reconstruct large portion of two parents genome with the IBD segments of relatives as the reference.