This paper presents a recursive secret sharing technique that distributes k-1 secrets of length b each into n shares such that each share is effectively of length (n/(k-1))*b and any k pieces suffice for reconstructing all the k-1 secrets. Since n/(k-1) is near the optimal factor of n/k, and can be chosen to be close to 1, the proposed technique is space efficient. Furthermore, each share is information theoretically secure, i.e. it does not depend on any unproven assumption of computational intractability. Such a recursive technique has potential applications in secure and reliable storage of information on the Web and in sensor networks.