Many process parameters in a wastewater treatment plant are expensive, difficult or even impossible to measure online, limiting the possibilities for efficient process monitoring and control. In this work, soft sensors were developed to provide on-line values for a number of parameters, primarily different fractions of phosphate (PO4 and total phosphorous), nitrogen (NO3, NH4 and total nitrogen), organic matter (COD) and suspended solids (TSS), at five different steps of the wastewater treatment process at the R&D-facility Hammarby Sjöstadsverk. The soft sensors were PLS (Partial Least Squares) models predicting the value of the hard-to-measure parameters based on easy-to-measure process parameters that were normally measured on-line or on acoustic data generated by acoustic sensors placed on the tanks of three of the five selected process steps. During a 13-day sampling campaign, data for the soft sensor development and validation were collected by laboratory analysis of the hard-to-measure parameters and combining them with corresponding 5 minute average values of the on-line parameters and the acoustic data. A majority of the soft sensors that were based on acoustic data had comparable or better performance than corresponding models using process data, indicating that data from acoustic sensors are of interest as input variables for soft sensors at WWTPs. The performance of the soft sensors varied significantly and some of them showed promising results. When removing the effect of the laboratory measurement error and the sampling error, 6 out of 26 soft sensor models had a so-called relative true prediction error less than 10% (NO3 in untreated water, COD, TSS and NO3 in the first bioreactor, NH4 in the last bioreactor and TSS in the membrane bioreactor). In combination with the proposed actions for further improvement of the models, the results suggest that soft sensors, that in many cases preferably could be based on acoustic data, is a possible approach to provide WWTPs with on-line process data.