The molten globule is generally 10-20% larger (by hydrodynamic radius) then the natively folded protein

Hydrodynamic Radii of Native and Denatured Proteins Measured by Pulse Field

Gradient NMR Techniques

Biochemistry 1999, 38, 16424-16431

http://pubs.acs.org/doi/pdf/10.1021/bi991765q

Compaction during Protein Folding Studied by

Real-Time NMR Diffusion Experiments

J. Am. Chem. Soc. 2000, 122, 5887-5888 (and references within)

http://pubs.acs.org/doi/pdf/10.1021/ja994514d

Hi Rossana, as Jeffrey said, the molten globule state is a more expanded state than the native state assembly, and I don't refer to this as a conformational change but rather to a less compact state (loss of tertiary contacts) and there are many classical biophysical experiments to do, like binding to ANS or controlled proteolysis. Can you tell me the conditions in which you see a lower volume? This could help!

It's not impossible to have a lower overall radius of gyration in what otherwise one might see as a more open or random state. Certain helix rich proteins could for a start. I don't know if it will help but see page 587-592 in my old book "Introduction to Proteins and Protein Engineering (B Robson and J Garnier, 1988, Elsevier Press.) This describes polymer-statistic principles for engineering heteropolypetides to have specific  dimensions in a random coil state including more contracted forms like a MG.  See also page 188 on the "liquid globule" and the rather similar  "Monomolecular Liquid Micelle" that I proposed in 1971, and see also the  references to our  penicillinase folding  there that appeared to contain such a state. There's a lot of related work before A. Wada and M Ohgushi in 1983 used the phrase  "molton globule", including from O B Ptitsyn, I think. Quite a lot of work then was experimental on radius of gyration via viscosity and related techniques, now rather forgotten, I think.

This could be possible if the protein has intrinsic disorder segments which may undergo disorder-order transition

For atomic force microscopy on a surface then there are issues of  denaturation

(PLoS One. 2013; 8(8): e73572.

Published online 2013 Aug 29. doi: 10.1371/journal.pone.0073572PMCID: PMC3756990Sequential Unfolding of Beta Helical Protein by Single-Molecule Atomic Force Microscopy

David Alsteens,#1,* Nicolas Martinez,#2,3 Marc Jamin,2,* and Françoise Jacob-Dubuisson4,*)

and of possible molten globule, random coil, or hybrid forms. There is a bit of an analogy with the polymer statistics of the "thrown string problem",  a collapse to the plane, that was never fully resolved. Robson, B. (1982)."A Comment on the Thrown String Problem" and submissions by others in, J. Royal. Statistical Soc. (B) 44, 136-138 Interesting to note also the following.

Protein Sci. 1998 Sep; 7(9): 2004–2011.

Compactness of the kinetic molten globule of bovine alpha-lactalbumin: a dynamic light scattering study.

K. Gast, D. Zirwer, M. Müller-Frohne, and G. Damaschun.

which distinguishes by volume a thermodynamically stabilized MG from the MG-like kinetic intermediate.

The following isn't strictly related to AFM as an observational technique, nor to the molten globule state per se, but this could be highly relevant seeing as molten globules are often defined by multiple criteria, and it's not clear which is being used here. 

The relationships between thermodynamically-defined volume changes of folding and the hydrodynamic radius are often unintuitive. (By "thermodynamically defined" I mean the apparent ∆V of folding in a constant-pressure experiment, so that the spontaneity is well-characterized by the Gibb's free energy.) "Protein only" techniques like AFM do not directly monitor the actual cooperative unit in the process of folding, which necessarily includes non-protein (solvent, ion, cofactor) atoms. I don't think the general explanation of the "protein volume paradox" — that the volume change of protein folding is often positive —is necessarily correct. 

Catherine Royer and peers recently explained this "paradox" in terms of volume voids that are present in the folded state and absent in the unfolded state. Their paper seems very good (linked below) but I'm unsure of the generality of the conclusion. Their simulations do not treat the pressure dependence of water structure itself, which I suspect is also relevant to the general phenomenon. (They do consider and eliminate this as a possibility, at least rhetorically, and they appeal to prior results that had significant amounts of non-protein cosolutes.)

Also pay attention to the group of A. Joshua Wand, who have recently made some discoveries regarding this phenomena in the context of their (pet) reverse-phase micellar systems.

http://www.pnas.org/content/109/18/6945.full

Thank everyone for the answers and comments. It took me some days to come back since my connection from Cuba is not always Ok. will try downloading the papers you suggest and come later with some information about the conditions we used,the protein, etc.