The traditional classification of infant attachment described three distinct types (Ainsworth et al. 1978): Secure (B), Insecure-avoidant (A), and Insecure-resistant (C). Research shows that the quality of infant attachment reflects the child's history of interaction with their primary caregiver and, therefore, maternal sensitivity and appropriateness of maternal responses during the first year of life has been found to predict infant attachment. In this study Nonlinear Dynamic Systems (NDS) approach was applied to broaden the study of maternal sensitivity into the overall temporal organization of mother-infant relationship exchanges. The study focuses on understanding the differences between secure and insecure attached children by applying NDS in two temporal scales: real time and a developmental scale, with the notions of 'flexibility' and 'self-organization', respectively. Infants, classified as securely or insecurely attached at 15 months, had free-play situations with their mothers, at 6 and 12 months of age, videotaped and coded in real time. Results showed that at 6 months dyads from the B group, compared to the non-B group, showed higher flexibility through several NDS indices derived from the State-Space Grid method (SSG). The dyads at 12 months did not show differences in those indices. Moreover, B group showed self-organization by decreasing the number of attractors, from 6 to 12 months of infant's age, in contrast with A and C groups that either showed less self-organization, by increasing the number of attractors, or stayed basically as they were at 6 months. Furthermore, the B group showed an increase in the proportion of attractors with higher values from time 1 to time 2, in contrast to the non-B groups. Findings provide some grounds for using a SSG approach to deepen the construct of maternal sensitivity in dyadic terms.