Abstract
Introduction
Stroke survivors often fall during walking. To reduce fall risk, gait testing and training with avoidance of virtual obstacles is gaining popularity. However, it is unknown whether and how virtual obstacle crossing is associated with fall risk.

Aim
The present study assessed whether obstacle crossing characteristics are reliable and assessed differences in stroke survivors who prospectively experienced falls or no falls.

Method
We recruited twenty-nine community dwelling chronic stroke survivors. Participants crossed five virtual obstacles with increasing lengths. After a break, the test was repeated to assess test-retest reliability. For each obstacle length and trial, we determined; success rate, leading limb preference, pre and post obstacle distance, margins of stability, toe clearance, and crossing step length and speed. Subsequently, fall incidence was monitored using a fall calendar and monthly phone calls over a six-month period.

Results
Test-retest reliability was poor, but improved with increasing obstacle-length. Twelve participants reported at least one fall. No association of fall incidence with any of the obstacle crossing characteristics was found.

Discussion
Given the absence of height of the virtual obstacles, obstacle avoidance may have been relatively easy, allowing participants to cross obstacles in multiple ways, increasing variability of crossing characteristics and reducing the association with fall risk.

Conclusion
These finding cast some doubt on current protocols for testing and training of obstacle avoidance in stroke rehabilitation.