To study homicide trends by gender and age, 1979-1994, Sao Paulo, Brazil, cubic polynomials were used to determine the best model for adjusting to time trends in homicide mortality rates by age and gender in the city of Sao Paulo, Brazil, 1979-1994. The model best adjusted to each group was selected considering the regression coefficient (beta), R² value, residual analysis, and model's simplicity. The results show linear growth for total rates and rates by gender due to behavior of rates in the 20-29 and 30-39 year age groups. The reciprocal model adjusted best to rates for the 40-49 and 50-59 year age groups, while rates for adolescents followed the multiplicative model. There was no significant relationship between homicide rates and time for the remaining groups (under 10 and over 59 years). Rates for males were considerably higher in all age groups. The remarkably steady growth in homicide rates among adolescents and young adults is consistent with trends observed in other urban areas in developing and developed countries and denotes deteriorating living conditions and increased poverty.