Experiments on particles' motion in living cells show that it is often subdiffusive. This subdiffusion may be due to trapping, percolation-like structures, or viscoelatic behavior of the medium. While the models based on trapping (leading to continuous-time random walks) can easily be distinguished from the rest by testing their non-ergodicity, the latter two cases are harder to distinguish. We propose a statistical test for distinguishing between these two based on the space-filling properties of trajectories, and prove its feasibility and specificity using synthetic data. We moreover present a flow-chart for making a decision on a type of subdiffusion for a broader class of models.Experiments on particles' motion in living cells aimed on understanding molecular crowding [1][2][3][4] have unveiled that diffusion in such environments is often anomalous, i.e. the mean squared displacement (MSD) does not grow proportionally to time, x 2 (t) ∝ t , but follows a slower patternwith 0 < α < 1 (subdiffusion), and the nature of this anomaly has to be understood. Anomalous diffusion is not only apparent in biological systems, but is found in complex systems ranging from amorphous semiconductors [5], goeology [6], to turbulent systems [7]. There are several mathematical models leading to subdiffusion, corresponding to different physical assumptions about the structure and energy landscape of the system in which the subdiffusive motion takes place. Since one is mostly interested in the actual microscopic structure of the system, an important task is working out tests which allow for distinguishing between different models giving the same prediction for the MSD. The three most popular models which might be pertinent to explaining subdiffusion in cells are:(i) continuous time random walk (CTRW), a mathematical model which is physically realized in systems with traps, i.e. binding sites of different energetic depths, a case pertinent to energetic disorder, (ii) diffusion on fractal structures, as exemplified by percolation, a situation pertinent to structural disorder, and (iii) fractional Brownian motion [8] (fBm), a Gaussian process with stationary increments which satisfies the following statistical properties: the process is symmetric, i.e x H (t) = 0 where x H (0) = 0, and the MSD scales as x 2 H (t) ∼ t 2H where H is the Hurst exponent. Note that H < 1/2 leads to subdiffusion, while H = 1/2 recovers Brownian motion. fBm physically corresponding to systems with predominating slow modes of motion and is realized in viscoelastic media as exemplified by polymers and polymer networks, where disorder does not play a leading role.Lastly, one has to discuss (iv) the time-dependent diffusion coefficient (TDDC) model -normal diffusion with a time-dependent diffusion coefficient, which is used to fit experimental results from, for example, FRAP (fluorescence recovery after photobleaching) experiments [9]. This model corresponds e.g. to a situation when the step rate is explicitly time-dependent, and does not have a clear physical ...