with participants' hand motion regardless of the task, while the correlation for some other features (e.g., warmth) depended on it. It can be speculated that for some perceptual features, the bottom-up effect is dominant in hand motion, while for others, the top-down effect is. Since preference, the main topic of this study, may reflect combinations of multiple perceptual aspects, it is highly likely that hand motion to judge preference will change as a result of both effects. We hypothesized that studying the strategy used (i.e., how an object is explored) to judge preference for an object's surface will reveal how humans extract information about tactile preference. An earlier study, by showing that the averaged velocity of finger motion did not have a correlation with pleasantness ratings, suggested that there is no bottom-up effect on hand motion 4 . However, that study restricted participants' touching mode to stroking only; therefore, the effect of the explorative strategy, i.e., the top-down effect, on preference judgment remains unknown.Capturing the features of the touching mode from complex patterns of hand motion is challenging, since the touching mode category is not explicit in most cases and participants may use more than one mode. It is known that how participants touch an object is time-varying even within a single trial, where the tactile stimulus is not changed and physical properties are constant. For example, participants sequentially have used a couple of touching modes, such as grasping and stroking, to judge an object's properties [13][14][15][16] . This view is also supported by a constructive approach showing that a robot arm could identify a tactile stimulus with higher accuracy by sequentially using multiple touching modes 17,18 . Other studies have shown that the averaged velocity and force tends to increase within a single trial 8,19 . Clearly, we need a better index than a categorical name or temporally averaged value to capture the temporal dynamics of hand motion.Here, we introduce nonlinear time series analysis as an effective tool for understanding the temporal dynamics of explorative hand motion, and try to reveal the links between the observed temporal dynamics of unrestricted hand motion and preference rating. Nonlinear time series analysis is a well-established method often used for analyses of other kinds of explorative motion such as eye movement during scans of radiographs 20 and wielding-hand motion during judgment of rod length 21 . One of the useful tools in the nonlinear time series analysis is the recurrence plot (RP) technique, which provides insight into the dynamics graphically 22 . Moreover, by calculating some characteristic values from RPs, such as DET, Lmax, and TREND as described below, we can quantitatively study the randomness, predictability, and stationarity of the time series, which is referred to as recurrence quantification analysis. Our main finding is that the index of stationarity (TREND) had a positive correlation with preference rating. We also fou...