Abstract We investigated the early stages of olivine crystal growth via in situ seeded experiments in a single plagioclase-hosted melt inclusion, using a heating stage microscope. Each experiment was subjected to a cooling ramp of 7800°C/h followed by an isothermal dwell at 19°C, 38°C, 57°C, 77°C, 96°C or 129°C of undercooling. The seeds (6–16 μm in diameter Ø) grew into large crystals (Ø 80–169 μm) in 3 to 30 min through the symmetrical development of tabular, skeletal, and dendritic overgrowths as the undercooling of the system increased. Time-resolved image processing and incremental measurements of the overgrowth thicknesses indicate up to three stages of crystal growth: an acceleration stage, a linear (constant growth rate) stage, and a deceleration stage. At the isotherm, the growth velocities reach a stable maximum that in all experiments corresponds to the period of linear growth. The highest linear values are measured at the $\left\{101\right\}$ interfaces, from 2.1 × 10−8 m/s at 19°C of undercooling to 4.8 × 10−7 m/s at 129°C of undercooling. Crystal growth is slower at other interfaces, in the ranges 1.9–7.6 × 10−8 m/s and 4.5 × 10−9 – 7.6 × 10−8 m/s for the $\left\{100\right\}$ and $\left\{001\right\}$ forms, respectively. Growth in the $<010>$ dimension appears limited to less than 2.4 × 10−8 m/s at 129°C of undercooling. We constrain the uncertainty on these growth velocities, which includes the environmental conditions (± 8.6°C on the nominal undercooling) and the measurements of crystal lengths (underestimated by <16% at most fast interfaces). A systematic and comprehensive review of 19 pre-existing datasets indicates that our linear growth velocities are faster than most growth rates determined at comparable undercoolings. Growth rates determined as half crystal lengths divided by total time are intrinsically low estimates of the true maximum, linear growth velocities, because the total time includes periods of slower or non-growth, and measured crystal dimensions are subject to projection foreshortening or truncation. These errors can lead to values that are several times to several orders of magnitude lower than the true maximum growth rates. This study completes and refines previously published data on the crystallization kinetics of olivine, highlighting the sensitivity of growth rates to specific environmental conditions and measurement methods. We emphasize the importance of symmetrical growth and true maximum growth velocities for interpreting olivine growth histories.