Since the late nineteenth century, physics has been puzzled by the timeasymmetry of the thermodynamic phenomena in the light of the apparent T-symmetry of the underlying laws of mechanics. However, a compelling solution to this puzzle has proved elusive. ln part, I argue, this can be attributed to a failure to distinguish two conceptions of the problem. According to one, the main focus of our attention is a timeasymmetric law-like generalisation. According to the other, it is a particular fact about the early universe. This paper aims (i) to distinguish these two different conceptions of the time-asymmetric explanandum in thermodynamics; (ii) to argue in favour of the latter; and (iii) to show that whichever we choose, our rational expectations about the thermodynamic behaviour of the future must depend on what we know about the past contrary to the common view, statistical arguments alone do not give us good reason to expect that entropy will always continue to increase.