Synaptic inputs onto single cortical neurons exhibit substantial diversity in their sensory-driven activity. What this diversity reflects is unclear, and appears counter-productive in generating selective somatic responses to specific stimuli. We propose that synaptic diversity supports computations that most effectively use information from upstream populations. To test this idea, we directly compare in silico synaptic weights for probabilistic decoder with in vivo synaptic input measurements. We define a decoder for a single sensory variable, orientation, that reads out the stimulus orientation from the responses of a realistic, hypothetical input population of neurons. Decoder synaptic weight variability was higher when upstream input populations consisted of noisy, correlated, and heterogeneous neurons, as is typically found in vivo. In addition, in silico weight diversity was necessary to accurately decode orientation. We then provide a straightforward mapping from the decoder weights to real excitatory synapses and found the diversity of decoder weights well-matched the functional heterogeneity of dendritic spine orientation tuning imaged in vivo. Our results indicate that synaptic weight diversity is a necessary component of information transmission and reframes studies of connectivity through the lens of probabilistic population codes. These results suggest that the mapping from synaptic input tuning to somatic selectivity may not be directly interpretable without considering input covariance and highlights the importance of population codes in pursuit of the cortical connectome.