Monitoring people during their daily activities has become a topic of great interest among the scientific community, largely due to the great advances in microelectronics, wearable devices, ubiquitous computing, or artificial intelligence. Among these advances, smartphones have arisen as a key technology to monitor people, thanks to their onboard sensors and their increasing computational capabilities. Most accessible smartphones are equipped with accelerometers, which has opened avenues to explore aspects based on human mobility. For example, linear accelerations present patterns during human walking cycle allowing for analyzing the gait. Despite identifying those patterns with smartphones is feasible, some restrictions must be considered, for example, the device position in the human body or battery consumption.In this paper, we compare the performance of time domain algorithms for step detection considering: (1) varying sampling frequency of the data; (2) different positions of the device while users walk (i.e., hand, pocket and lower back); and (3) different paces during walking (i.e., low, normal and quick). The compared algorithms include vertical acceleration (VA), magnitude, energy and vertical acceleration using the inclination angle of the device (VAIA). Our results did not indicate considerable influence of the device position and user's pace in the performance of the algorithms based on magnitude, energy and VAIA. On the contrary, the lower performance was presented by the VA algorithm.