This longitudinal study examined flexibility in early mother-infant interaction at the age of approximately 6 months (N=30) and whether flexibility indices predicted (in) secure child attachment at 15 months. Dyadic flexibility was measured using dynamic systems-based modelling of patterns during mother-child free play in terms of NDS variables derived from SSG: the propensity to change states (dynamic flexibility), number of states visited (diversity) and predictability (dispersion). Results showed significant discriminant functions on the attachment type groups, A, B & C, for the total grid, which included verbal and non-verbal, and for the reciprocal verbal region. Specifically, the prediction outcomes seem to work better in total grid for A-dyads and in the reciprocal verbal region for B and C-dyads. Diversity emerged as the most relevant index in dyadic flexibility: A-dyads showed the least diversity, distinguished them from B-dyads in the verbal regions, (both the reciprocal and non-reciprocal, 'child verbal-mother non-verbal' one), and, from C-dyads in the reciprocal non-verbal region. A-dyads showed remarkably low activity in the regions involving child verbal behavior, showing that children who became avoidant attached at 15 months of age, were mostly silent at approximately 6 months, when they interacted with their mothers. Findings in this study contribute to advancing conceptually informed measurement of dyadic interaction to provide a new perspective on maternal sensitivity and early markers of child insecure/secure attachment.