In this paper, we improve previous preimage attacks on hash function HAS-160, which is standardized in Korea. We show that the last 68 steps out of 80 steps of HAS-160 can be attacked, while a previous attack works for only intermediate 52 steps. We also show that the first 67 steps of HAS-160 can be attacked. These attacks are based on the meet-in-the-middle attack, which is also used in the previous attack. Recently, various techniques of preimage attacks have been proposed on other hash functions. We show that these techniques can also be applied to HAS-160 and the number of attacked steps can be improved. For the attack on 68 steps, we first generate pseudo-preimages with a complexity of 2
, and then convert them to a preimage with a complexity of 2
. This attack uses a memory of 2
×7 words. To the best of our knowledge, attacking 68 steps is the best of all attacks on HAS-160 hash function.