Transforming growth factor-beta is a multifunctional cell-growth regulator and is a member of the TGF-beta superfamily of cytokines. Each monomer is 112 amino acids long and the mature active form is a 25 kDa homodimer. Recently, the crystal structure of TGF-beta2 has been determined independently in two laboratories [Daopin, Piez, Ogawa & Davies (1992). Science, 257, 369-373; Schlunegger & Grütter (1992). Nature (London), 358, 430-434] and subsequently refined to higher resolutions [Daopin, Li & Davies (1993). Proteins Struct. Funct. Genet. In the press; Schlunegger & Grütter (1993). J. Mol. Biol. In the press]. A detailed structural comparison shows that the two structures are nearly identical with the differences mostly located on the mobile regions of the molecule. The r.m.s. differences between the two structures are 0.10 A for 104 pairs of C(alpha) atoms, 0.15 A for 434 pairs of main-chain atoms, 0.33 A for 860 out of 890 pairs of protein atoms and a correlation of 90% between the temperature B factors of all protein atoms. Based on a comparison of the water molecules, a B value of 60.0 A(2) is recommended as the cut off for modeling new waters. The structural identity is striking because in one case the material was expressed in vivo in CHO cells whereas in the other case it was expressed in E. coli and had to be refolded in vitro. The overall coordinate errors are estimated to be 0.21 A from the Luzzati plot, 0.18 A from the sigma(A) plot, 0.24 A with Cruickshank's equations and 0.25 A using the empirical method of Perry & Stroud. These estimates are comparable to the r.m.s. structure superposition. The r.m.s. differences correlate very well with the crystallographic B values and the relation is best described with the Cruickshank formula. In addition to the estimation of an overall error, a new application of the Cruickshank formula is presented here to estimate the local errors.