Blocks with a duration of 62 TR (assuming a standard TR of 2 s this would be 2 minutes) are much longer than the usual settings. The 7 TR rest period in between is quite short. In theory it's sufficient for the task regressor to turn back to baseline, but right afterwards the next block starts. The design might be ineffective to properly estimate activations during the task relative to baseline, especially as block regressors are rather crude approximations for long durations, and as activations might be far from homogenous within look blocks (which might also hold for behavioral data, e.g. maybe subjects perform very well at the end of each trial, but very poor at the beginning). Note these are just reflections, it might be standard to go with similar settings for paradigms like yours (but they might nonetheless be problematic). 

Dear Joelle,

It depends on the purpose of the study.

1) You can model the trials separately, which allows to e.g. contrast trial 1 with trial 10. This is very flexible, as you make no assumptions about changes over time (in contrast to 2), see below). As this results in ten separate regressors it doesn't really make sense to take into account the behavioral data on single-subject level. However, you could e.g. set up contrasts like trial 10 > trial 1 for each of your subjects and forward this into a group model, with the difference in behavioral data as a regressor to test for associations between changes in performance and changes in activation.

2) You can go with a single task regressor. This allows to test for average visuomotor activations only, if you want to look at changes over time you would add a parametric modulator, reflecting the trial number. Depending on assumptions (linear changes, quadratic, ...) you could adjust the order of the PM accordingly. Note that this does not test for e.g. quadratic changes over time "in general", but only for a particular quadratic change over time, derived from the values entered  ^2 and convolved with the HRF. If you want to test for other relationships (other quadratic ones, exponential, ...) you would have to set up values yourself and go with first order (which means the values are unaltered, or only mean-centered, thus testing for a "linear" relationship with the e.g. exponentially scaled values). You can set up several different PMs based on different values, e.g. one first-order PM based on 1:1:10, testing for linear changes over time, and another first-order PM based on exp(n) with n=1:1:10, testing for exponential changes over time. Again note that 1) this "exponential function" might not be the one you're interested in, maybe you would rather want to look at exp(1-n) with n=1:1:10 and 2) make sure whether you want to use orthogonalization (usually we want to at least orthogonalize the PM regressors onto the unmodulated task regressor, but possibly also serially orthogonalize the 2nd PM onto the 1st one). In a similar manner you could also include behavioral data as a 2nd or 3rd PM. Similar to 1) you could also leave behavioral data unconsidered during single-subject models but test for relationships between the PMs and some parameters derived from the behavioral data (e.g. beta images reflecting linear activation changes over time and behavioral score reflecting linear changes over time).

Best

Helmut