Various application scenarios of a smartphone sometimes require one-handed and/or eyes-free interaction. Tilt-based interfaces have the potential to meet these requirements. Taking multiple application scenarios into account, we conducted an experiment to systematically investigate human ability in controlling tilt input of a mobile phone. Three visual feedback levels, i.e., fully visual feedback (FV ), partially visual feedback (PV ), and no visual feedback (NV ), were investigated. Under the NV condition, the participants performed a task using an eyes-free method. The results revealed that trials were performed the fastest but were the most error-prone under the NV condition. The participants could easily distinguish 4 tilt orientation levels (TOLs) and 2 tilt magnitude levels (TMLs) or 8 TOLs and 2 TMLs under the NV condition with tolerance of an error rate 10% or 15%, respectively. We also found out that the participants' abilities to control tilt input were related to tilt orientation directions. The results have some implications for non-visual interface designs using tilt as primitive input.