Model functions were fitted (correlation coefficient [r], from 0.9258 to 0.9999) to the experimental microbial growth data in orange juice stored at 0, 4, 10 and 15C. The models predicted a microbial population of about 5.3 log cfu/mL after 15.5 days (0C) and that it would take 12.7 days to reach a microbial population of 6.0 log cfu/mL (4C). The first derivative of the microbial growth model function (10C) indicated a growth rate of about 0.5 log cfu/mL/day (day 1). The lag phase period at 0 and 4C was 9.5 and 2.8 days, respectively. The shelf life of orange juice (time to reach 6 log cfu/mL) determined experimentally was 17.9, 12.7 and 3 days and 10 h at 0, 4, 10 and 15C, respectively. Three shelf‐life predictive models were developed (r = from 0.9436 to 0.9999). The predictions of shelf life by the “algebraic” model were identical to those determined experimentally. This was also true for the relative rate of spoilage (RRS) model at 0 and 15C. On the other hand, the RRS model underestimated the shelf life of orange juice at 4 and 10C. The modified exponential model overestimated (at 0, 10 and 15C) and underestimated (at 4C) the shelf life of orange juice. An Arrhenius “split” plot with a break point at 7.6C depicted the effect of temperature on the rate of development of microbial generations. The absolute value of the slope of each portion, of the Arrhenius plot, above and below the break point was 12.5 and 4.5, respectively. The slope of the Arrhenius slope for the temperature range below the break point (7.6C) indicated a slower microbial growth.